WritersLogic verifiable AI agents

Abandoned Compound & Uncommercialized Trial Miner — Finds shelved drug and biologic candidates worth a second look. Give a disease or target (mines ClinicalTrials.gov for terminated, withdrawn and suspended trials and for completed Phase 1/2 non-industry trials with no later industry Phase 2/3/4 trial), a compound you already have in mind (drug name or ChEMBL id), a trial (NCT id), or a desired material property (literature mode, no trials). Every

Signed acceptance receipt: which deterministic tests a delivered output passed — Runs your deterministic acceptance tests (equals, exists, absent, type, contains, length and number ranges, one-of, sha256, a JSON Schema subset) against a sub-agent's or vendor's JSON output at JSON Pointer paths, and returns per-test results plus a hub-signed receipt over the output hash, the tests hash and the results. Use it before paying for delegated work or releasing the next step. It certi

ADA / WCAG Lawsuit Shield & DOM Auto-Patcher — Returns a FAIL / REVIEW / PASS verdict for a page's HTML (pasted or fetched from a URL) against the WCAG failures most cited in ADA demand letters: alt text, form labels, unnamed buttons and links, lang, headings, landmarks, skip links, zoom. Each issue carries a CSS selector, source line and fix, with a top-five list, a SARIF 2.1.0 log, a litigation risk score, per-element patched HTML and an int

Age in complete years on a date and the exact dates a minimum or maximum age is reached (leap-day births) — For up to 100 people, computes age in complete years and months on a caller asOf date, the date they reach each minimum age (the birthday) and the first day they exceed each maximum age, and checks min/max age requirements; 29 February births need a declared common-year rule (28 February or 1 March) because jurisdictions differ. Built for an agent checking eligibility (18+, under 65, 21 to buy). I

AI-BOM License Chain Resolver for Hugging Face Models — Walks the Hugging Face lineage of up to 10 models (base_model chain, training datasets; at most 30 nodes, 4 generations) and applies a curated, hash-pinned table of license clauses (Apache-2.0, MIT, CC-BY, CC-BY-NC, CC-BY-SA, OpenRAIL variants, Llama 2 / 3 / 3.1 / 3.2 / 3.3, Gemma, Mistral Research License) to each intended use: commercial or not, API, weights or on-prem, fine-tuning, monthly acti

Allergen flags in an ingredient list (EU Annex II, US FALCPA + FASTER Act) — Scans an ingredient list for terms tied to the 14 allergens of EU Regulation 1169/2011 Annex II and the 9 US major food allergens (FALCPA plus sesame under the FASTER Act), for an agent or buyer checking a label or recipe. Returns each matched span with its character offsets, the allergen, the regime that lists it, and whether it sits in a 'may contain' clause. Pass the allergens a label declares

Split a money amount into parts that sum exactly to the total — Splits one signed amount into up to 500 parts by weights (or equal parts), in integer minor units of its currency, for an agent dividing a bill, a refund, a fee or a tax. Each share is rounded under the declared mode, then the leftover minor units are assigned by the declared remainder policy, so the parts always sum to the total exactly. Negative totals (refunds) are supported; zero-weight parts

Does this appointment fit: opening hours in the venue's zone, closed days, existing bookings with buffers, minimum notice — Checks one proposed appointment (instant + duration) against a venue's local opening hours per weekday, closed dates or a named holiday calendar, up to 500 existing bookings with a buffer on each side, and a minimum notice measured from a caller-supplied asOf; every violated rule is returned as a witness with the conflicting booking or hour block. Local hours are resolved in the venue's IANA zone

Does a change to the approved thing invalidate the approval under a materiality policy? — Compares what a human or system approved with what is about to be executed (a quote, order, plan, diff summary: any JSON) under a caller-supplied materiality policy: per-path rules exact / ignore / numeric tolerance (absolute or percent) / no-increase / no-decrease, with the longest matching path winning and a default for unlisted paths. Also covers quote revisions (approved revision number must e

Cross-Platform Arbitrage & Resolution Rules Lawyer — Finds fine-print divergences between two prediction markets (e.g. Polymarket vs Kalshi), prices basis risk after fees, and returns verdict.action ARBITRAGE_SAFE / LOW_ANNUALIZED_YIELD / BASIS_RISK_TRAP / NO_EDGE / INSUFFICIENT_EVIDENCE with edge in bps, the probability both legs lose, days to resolution, annualized return and Kelly hedge size capped by top-of-book depth. Each market can be given a

Archive extraction preflight: zip-slip, duplicate destinations, symlink conflicts — Checks a manifest of up to 2000 archive entries (path, type, link target, sizes) before extraction, for an agent that unpacks zip, tar or OOXML files: ../ traversal (also through backslashes), absolute, drive-letter and UNC paths, entries that write through a symlink, symlink and hardlink targets that leave the root, duplicate destinations, case-only and NFC/NFD collisions, Windows reserved names,

Which derived artifacts must be rechecked after these inputs changed? (bounded dependency graph) — Given a dependency graph of artifacts (each lists the artifact ids it was derived from) and the ids that changed, returns every artifact that transitively depends on a change, in an order safe to recompute (inputs before dependents), and the artifacts that are unaffected. A cycle or an unknown input id is a defect in the graph and returns fail with the cycle path. Changed ids that are unknown fail

arXiv-to-Production Code & Hyperparameter Compiler — Fetches an arXiv paper's full text (ar5iv), selects method, pseudocode, hyperparameter tables and ablations, and returns a core-algorithm summary, hyperparameters each with a source pointer and a verbatim quote verified against the text, ablation warnings with evidence sentences, and one self-contained TypeScript or Python file that is compile/syntax-checked and literal-checked against the extract

Find assumptions that disagree across agents (currency, timezone, version, customer id, ...) — Takes assumptions that several agents or steps hold about the same fact (key, value, who) and reports every key where the values differ, naming the conflicting pairs and whether they differ only by case or whitespace (a formatting difference) or in substance (USD vs EUR, a different id, a different version). Optionally requires each listed agent to state each required key; a missing statement is u

Check that what a message says is attached matches the attachments actually supplied — Compares a message's wording with its attachment list: 'see attached' with nothing attached, a named file (report.pdf) that is not in the list, a kind of file mentioned (the spreadsheet, a screenshot, the PDF) that no attachment matches by extension or MIME type, and a stated count ('two files') that differs from the list; attachments the text never mentions are listed. Mentions inside a negation

Forensic AI-trajectory & prose craft auditor — Deterministic prose forensics: word echoes within 50 words, repeated 3-6-grams, sentence-starter runs, hedge/filter/passive/nominalization density, burstiness, MATTR, Flesch-Kincaid, vocabulary fingerprints, AI syntactic tics (Not X. Y. pivots, verbless fragments, autonomous body parts), optional revision-history trajectory, and an editorial humanization map whose quotes are verified verbatim. Eve

OnRecord statutory clock & redaction auditor — Audits an agency's public-records response and leads with a decision (late, redaction problems, incomplete, pending or no violation found). Computes the statutory response deadline and days late (California CPRA, with the court-day calendar and § 7922.535 extension rules), flags redactions without an exemption citation, passages redacted in one document but readable in another, referenced attachme

Bid coverage: uncovered lines, substitutions, partial quantities and conditional bundles — Checks which required lines of a tender or RFQ are covered by the bids received (up to 200 lines, 30 bidders, 2000 offers), for an agent comparing responses. A line is covered when firm offers (exact item, no condition, no bundle) sum to the required quantity; substitute offers, offers bound to a bundle (all-or-nothing across the lines in that bundle) and offers with a stated condition are reporte

Which cancellation or change tier applies at this moment, and what does it cost? — Given a booking start, the moment of change, the booking amount and a tiered policy whose deadlines are either elapsed hours before start ('48 hours') or local calendar days at a local time ('18:00 the day before' in the venue's zone, DST-correct), returns the applicable tier, its deadline instants, the penalty in minor units under your stated rounding, and when the next tier begins; checks an ass

Business-day deadline (clear days, cutoffs, DST) on a named or caller calendar — Computes the date (and optionally the instant) a business-day deadline falls on, for an agent drafting or checking a contract, notice, SLA or partner response window: '10 clear business days' (both endpoints excluded) or 'within 10 business days' (end day counted), forward or backward, with an after-cutoff roll, caller or named holiday calendar and a DST-aware deadline time. Named calendars (US fe

Check that a result-cache key contains every input that changes the result, and nothing volatile — Given the inputs of a cached computation (each declared as influencing the output or not, optionally volatile) and the list of inputs the cache key is built from, finds influential inputs missing from the key (stale or wrong hits), volatile inputs inside it (the cache never hits), unknown-influence inputs left out, and, for private results, a missing tenant dimension (cross-tenant leak). When valu

Convert dates between Gregorian and Hijri (named variants), Hebrew, Persian, Ethiopian; Western and Orthodox Easter — Converts a Gregorian date (or a Hebrew, Persian, Ethiopian or Hijri date) to the other calendars via the runtime's ICU, always naming the Hijri variant (Umm al-Qura, tabular civil, tabular astronomical-epoch, ICU astronomical) and flagging when variants disagree on the day; with mode easter, returns Western (Gregorian computus) and Orthodox (Julian computus, with its Gregorian equivalent) Easter a

Calibration certificate vs intended use: range, uncertainty and validity — Checks whether an instrument's calibration certificate covers an intended measurement, for an agent about to rely on a reading or accept a calibrated device. Three checks: the intended range [min, max] lies inside the calibrated ranges (touching or overlapping ranges are joined); the stated expanded uncertainty of every calibrated range used is no larger than the required uncertainty; the calibrat

Decode EVM calldata, return data, revert data or event logs against an ABI — Decodes hex calldata (selector plus arguments), function return data, revert data or an event log against a caller-supplied ABI using ethers, for agents inspecting a transaction before signing or after it fails. Handles tuples, nested and dynamic arrays, bytes and strings, and the built-in Error(string) and Panic(uint256) reverts; a strict re-encode check flags trailing bytes, dirty padding and no

Pack items into the smallest fitting carton (bounded heuristic) — Checks whether up to 50 box-shaped items (axis-aligned, 90-degree rotations, optional this-side-up lock, optional weights) fit into any of up to 20 candidate cartons, and recommends the smallest-volume carton for which a packing was found, with coordinates for every item. For shipping and fulfilment agents sizing a parcel. The packing is a greedy extreme-point HEURISTIC with a fixed step cap, not

Cash rounding to the smallest payable coin (CHF, CAD, SEK, AUD or a caller rule) — Rounds up to 100 final cash totals to the smallest payable increment, for an agent computing what a customer actually pays in cash or the change owed. Built-in rules exist only for currencies with an official source: CHF 0.05 (Münzverordnung SR 941.101), CAD 0.05 (Government of Canada), SEK 1 (Lag 1970:1029), AUD 0.05 (Consumer Protection WA); any other currency needs a caller increment. Rounds th

Causal Claim Referee — Referees a causal claim against an observational table (JSON rows, CSV text, or a published contingency table with counts) and leads with a verdict, confidence, whether the claim holds, the crude and adjusted effects with bootstrap intervals, and a concrete next step. Underneath: crude and stratified effects, Mantel-Haenszel pooling, Simpson's-paradox detection, d-separation and backdoor/front-doo

E-Commerce Chargeback Representment Builder — Tells a merchant whether to fight or accept a card dispute: scores win probability against the network rubric for the reason code (Visa, Mastercard, Amex, Discover), runs deterministic Visa CE3.0 eligibility and invalid-dispute checks, computes expected recovery net of the contest fee and days to the deadline, and returns a processor-ready submission (summary under 4,000 characters, evidence field

Chat completion, deep tier (12k in, 2k out) — OpenAI-style chat completion served by the hub's provider router (strongest available model, up to 12000 input tokens and 2048 output tokens per call), with a signed receipt that binds the provider, model, token usage and the hashes of your messages and the reply. Supports optional strict json_schema output returned both parsed and as a string. Does not stream, call tools or functions, accept imag

Chat completion, fast tier (8k in, 1k out) — OpenAI-style chat completion served by the hub's provider router (small, fast model, up to 8000 input tokens and 1024 output tokens per call), with a signed receipt that binds the provider, model, token usage and the hashes of your messages and the reply. Supports optional strict json_schema output returned both parsed and as a string. Does not stream, call tools or functions, accept images, or us

Chat completion, standard tier (12k in, 2k out) — OpenAI-style chat completion served by the hub's provider router (mid-size general model, up to 12000 input tokens and 2048 output tokens per call), with a signed receipt that binds the provider, model, token usage and the hashes of your messages and the reply. Supports optional strict json_schema output returned both parsed and as a string. Does not stream, call tools or functions, accept images,

Identifier check digits: IBAN, ISBN, GTIN, VIN, ISIN, LEI and more, in a batch — Validates the format and check digit of up to 100 identifiers per call, for an agent about to put an IBAN, ISBN, GTIN/EAN/UPC, container number, VIN, ISIN, CUSIP, SEDOL, LEI, IMO number, BIC or Luhn number into a payment, order or record. Each item is {kind, value}; with kind omitted the engine proposes candidate kinds and answers unknown if two kinds both validate. Reports the normalized form (lo

Chess move legality, SAN/UCI conversion and next FEN — Checks whether a move (or a sequence of moves) is legal from a FEN position under the FIDE Laws of Chess, for agents that play, annotate or verify chess games. Accepts UCI or SAN, returns the normalized move, the new FEN, check / checkmate / stalemate / insufficient-material flags and, if asked, every legal move; an illegal move names the reason (castling_through_check, ep_exposes_king, moves_into

Chord check: identify, spell, key membership, transpose — Names a chord from its notes, spells a chord symbol into notes, tests whether a chord is diatonic to a key, or transposes a chord symbol, using pitch-class arithmetic over a fixed table of 21 chord qualities, for agents handling lead sheets and music metadata. Rule-based and labelled: when several readings are equally valid (C E G A is C6 or Am7/C) or no quality in the table matches, it returns un

Split text into chunks under a hard token cap without cutting a table, sentence or code fence — Plans chunk boundaries for text an agent must feed to a model in pieces: every chunk is at most maxTokens in the named tokenizer (o200k_base or cl100k_base, verified by counting each finished chunk), boundaries fall only between sentences, blocks, whole Markdown or HTML tables and whole code fences, headings stay with the text that follows, and the chunks tile the input exactly (concatenating the

CIDR math: IPv4 and IPv6 overlap, adjacency, merge, subtract, membership — Computes exact CIDR relations for agents planning networks, firewall rules or allow-lists: for up to 200 IPv4/IPv6 blocks it reports which pairs are equal, nested, partially overlapping or merely adjacent, the minimal merged CIDR list, a subtraction (blocks minus exclusions), address membership with the most specific match, or each block's first/last address, count and host-bits-set errors. It doe

Citation Audit — Bibliography in, evidence audit out: reference-verifier checks every reference (fabricated, mis-cited, retracted, corrected); the top DOI-bearing references (corrections first, then the most cited in the manuscript) get a replication-auditor pass on the sentence that cites them; when a target is contested, paper-contradictions compares it with the studies against it. The contradiction check is opt

Locate quoted text in a source with exact offsets, or classify it as normalized, elided, paraphrased or absent — Finds each quote an agent attributes to a source text and returns where it sits: exact offsets in UTF-16, code points, UTF-8 bytes and graphemes, so the span slices back to the matched text. Tries an exact match, then a match tolerant of whitespace, curly vs straight quotes, dash types, ellipses, invisible characters and Unicode forms (NFC/NFD, ligatures), naming which normalizations were needed,

Prediction-Market Clarification Impact Pricer — Prices a prediction-market rules clarification as a trade. Given an open Polymarket market (slug, URL or conditionId; Kalshi ticker supported), reads the posted clarification (Additional context) from the UMA adapter on Polygon, or takes your clarificationText to price a pending or hypothetical one. Finds the real-world outcomes whose resolution changed, quotes each changed clause verbatim, shifts

Rule-based clause tagging (obligation, prohibition, permission, condition, topic) — Tags each contract-style clause you supply with the modal force it expresses (obligation, prohibition, permission, recommendation) and topic cues (definition, condition, termination, payment, confidentiality, liability, governing law, notice, assignment), with the matched cue and its offset. For an agent that must triage many clauses before a human or model reads them. Fixed cue phrases only (plus

Certificate of analysis vs specification: pass, fail or indeterminate per clause — Checks the values on a certificate of analysis or test report against specification limits clause by clause (up to 200 clauses), for an agent accepting a lot or a supplier's material. Each clause has a parameter, an optional min and max and a unit; each result has a value, an optional reported uncertainty and an optional less-than or greater-than qualifier ("<0.01"). rule simple (default): the rep

Company identity: name, domain, LEI, VAT, CIK or ticker to one verified legal entity — Resolves a company name, web domain, LEI, EU VAT number, SEC CIK or stock ticker to one legal entity using official free registries in a single call: GLEIF (legal name, other names, addresses, jurisdiction, registry id, direct and ultimate parent), the SEC ticker table (CIK, ticker), the EU VIES VAT service, and Companies House officers when the operator has configured a key. Deterministic matchin

Check that every write step in a multi-step plan has an undo step, and give the rollback order — For an agent plan made of steps with declared effects (read or write), checks every write in the plan, run or not, has a compensation step that exists, is itself a write, is not its own compensation, and that the dependencies have no cycle; flags writes declared irreversible; and, given which steps have completed, returns the compensations to run in reverse completion order. Does not judge whether

SaaS Pricing & Competitor Battlecard Spy — Reads up to 3 competitor sites (homepage and pricing page, fetched in parallel) and returns a deal battlecard in under a minute: a verdict headline with confidence, pricing tiers whose amounts are checked against the page text, a scenario cost comparison computed in code, hidden fees and limits, landmine questions, objection handling, wedge features and a talk track. If a site or model step fails

Rank the most compressible (cliched) phrases in a chapter — Finds the phrases in a passage that compress most against a reference corpus of generic prose, so a writer can target the low-entropy spans instead of guessing. For each sentence (default) or sliding n-gram, computes the marginal compressed-byte contribution per character with zlib level 9 and a built-in reference corpus of common narrative clichés; the lowest-scoring spans are the most predictabl

Migrate a v1 config to a v2 schema with an explicit mapping, and list what was dropped or defaulted — Applies a declared mapping (from pointer, to pointer, optional transform such as string-to-number or a value map) to a v1 JSON config, carries over unmapped keys the v2 schema declares, fills schema defaults, validates the result against the v2 JSON Schema 2020-12 and returns the migrated config plus three exact lists: dropped (v1 values that landed nowhere, with their JSON pointers), defaulted (v

Classify a confirmation message as submitted, pending, reserved or completed from caller or default phrases — Reads a confirmation page, email or API message and decides which state it states: submitted, pending, reserved or completed, using phrase lists the caller supplies (preferred, because wording is provider-specific) plus an optional default English list. Matches are whole-word, case- and typography-insensitive, returned with offsets in UTF-16, code points, UTF-8 bytes and graphemes, and a match ins

Check that earlier constraints (numbers, dates, codes, quoted terms, negations) survive into a later plan or summary — Takes the constraints an agent was given (budget caps, deadlines, identifiers, prohibitions) and a later artifact (plan, summary, handoff, tool arguments) and reports for each constraint whether its concrete atoms still appear: numbers with their unit or currency, ISO dates, codes, quoted terms, or caller-listed anchor strings, with offsets. For a negated constraint it also reports whether the clo

Pick context blocks that fit a token budget, keeping mandatory blocks and their dependencies — Chooses which context items an agent should put into a prompt without exceeding a token budget: every mandatory item and everything it depends on is always included, then optional items by descending priority (each with the dependencies it pulls in) while they fit. Token counts come from the caller or from o200k_base / cl100k_base; any other tokenizer name returns unsupported (there is no Claude t

Smart Contract Auditor — Source-level security review of Solidity: pass/warn/fail verdict, 0-100 risk score, severity-ranked findings with exact file:line, exploit scenario and a fix, plus a SARIF 2.1.0 log. Accepts raw code, a list of files, or an on-chain address (verified source is read from Sourcify). A deterministic detector pass (reentrancy ordering, tx.origin, delegatecall, unchecked calls, weak randomness, pre-0.8

Dark Pattern Contract & Subscription Exit Lawyer — Paste a contract or Terms URL and get the decision first: send notice by a computed date, or the window has passed. Quotes auto-renewal, notice-window and early-termination traps verbatim, computes the renewal and last-notice dates in code, estimates the renewal cost you avoid from your fee, maps the exit to ROSCA, state automatic-renewal laws and EU/UK consumer law with links, and drafts a notice

WCAG 2.2 contrast check with the minimal OKLCH lightness fix and colour-blind pair check — Checks up to 50 foreground/background hex pairs against a WCAG 2.2 contrast threshold (4.5 AA text, 3 large text and UI, 7 AAA, or your own) on the unrounded ratio, so 4.499 fails and exactly 4.5 passes, and for a failing pair returns the closest colour by OKLCH lightness (chroma and hue held, gamut-clipped, 8-bit) that meets it. Optionally simulates protanomaly, deuteranomaly and tritanomaly (Mac

Choose which old messages to drop to hit a token target without touching the cached prompt prefix — Given a conversation, the index of the last message inside the provider's cached prefix, and a token target, returns which messages to drop: whole tool call plus tool result pairs (so no result is left without its call) and messages the caller marked droppable, oldest first or largest first, never inside the cached prefix, never the most recent keepRecent messages, never pinned ones. Everything up

CORS preflight decision from Fetch spec — Takes a browser request shape (origin, method, headers, credentials flag) and a server CORS policy (allow-origin, allow-methods, allow-headers, allow-credentials, expose-headers, max-age) and returns the exact decision a browser would make per the Fetch Living Standard §3.2 and §4.10: whether a preflight OPTIONS is required, whether the main request is allowed, which response headers the script ca

DeFi Cover Claim Adjudicator — Predicts whether a DeFi cover claim (Nexus Mutual or any provider's wording) will be approved and for how much. Reads the wording, postmortems and the claimant's transactions, quantifies net token loss with Chainlink prices at the loss block, maps facts to covered events, exclusions and conditions with verbatim verified clauses, cites similar past claims, and returns verdict.action (FILE / FILE_WI

CRA Exploited-Vulnerability Evidence and Art. 14 Deadline Clock — Takes a CycloneDX or SPDX JSON SBOM (or a purl list), asks OSV.dev for the advisories that affect each exact component version (OSV evaluates the version ranges per ecosystem), then checks each CVE against the CISA Known Exploited Vulnerabilities catalog and the ENISA EUVD exploited-vulnerability index, and adds FIRST EPSS scores. Returns a per-component verdict, the evidence with source URLs and

EU CRA Readiness Check — Repository or manifests in, EU Cyber Resilience Act Art. 14 position out: dependency-advisor resolves every declared package to its locked version (deterministic, no model), those versions go to cra-exploit-clock as package URLs (OSV, CISA KEV, EPSS, EUVD), and each actively exploited component is joined back to the manifest that declares it with the fixed versions and the advisor's findings. Repo

Do licences, certificates or insurance policies cover a required period without gaps? — Given a required date range (a contract term, a project, a shipment window) and up to 200 credentials each valid from a date to a date (both inclusive), reports the merged covered ranges, every uncovered gap with its first and last day and length, and whether each checkpoint date is covered; a credential ending the day before the next starts is contiguous. Built for an agent verifying insurance, l

Cron parse, humanize and next N runs — Parses a POSIX 5-field or Quartz 6/7-field cron expression (plus macros @yearly/@monthly/@weekly/@daily/@hourly), returns a plain-English summary and the next N scheduled runs in a given IANA timezone, and reports if the expression never fires or lands on a DST-ambiguous wall-clock. Handles step syntax, named months and days, Quartz '?'/L/W/# extensions, and the POSIX dom-OR-dow rule when both fie

Check caller-declared rules across fields of a JSON record: end after start, conditional required, exactly one of — Evaluates up to 100 rules you declare over one JSON document and reports, per rule, pass, fail or unknown with the JSON pointer of every violating field and the actual value class (absent, null, empty string, value). Rule kinds: compare (>, >=, <, <=, ==, != on numbers, exact decimals, RFC 3339 dates or date-times, strings), required-if, forbidden-if, exactly-one, at-most-one, at-least-one, all-or

Find dangling and duplicated cross-references (Section 3.2, Table 4, Figure 2, [7]) in a document — Scans a document for the targets it defines (numbered headings, Table/Figure/Listing/Equation captions at line start, Appendix/Annex headings, [n] bibliography entries, [^id] footnote definitions) and the references it makes to them ('see Section 3.2', 'Tables 2 and 4', 'Figures 3 to 5', '[7]', '[^note]'), and reports references with no target, targets defined twice, and targets never referenced,

Crosstalk epistemic swarm & falsification engine — Stress-tests a research question instead of answering it once. Classifies the domain, generates up to 3 independent hypotheses, attacks each with an adversarial falsification pass, checks live literature on arXiv and OpenAlex (plus any papers you supply by arXiv id or DOI; every record content-addressed by sha256), then synthesizes a calibrated answer. The output leads with a headline (status, ans

Content-Security-Policy audit — Parses a Content-Security-Policy header or value and reports every clause that weakens it: 'unsafe-inline' without nonces or hashes, 'unsafe-eval', wildcard hosts, data:/blob: in script-src, scheme-only https:, duplicate directives (second is discarded), missing base-uri or object-src 'none', open frame-ancestors or form-action, and the strict-dynamic cases that silently ignore host sources

Detect a CSV file's delimiter, quote, header row and broken header names — Detects the delimiter (comma, semicolon, tab, pipe), quote character, line terminator, BOM, header row offset (preamble lines above the header) and whether a header exists in CSV text, parsed per RFC 4180 so quoted delimiters and CRLF inside quoted fields are handled. Returns every candidate dialect with its score, each ragged row with its physical line, and header repairs (empty, duplicate or BOM

Will this CSV survive a typed round trip and a spreadsheet: leading zeros, big ids, dates, formulas, empty vs null — Parses CSV text (RFC 4180, caller dialect) against a column type map and reports, per cell with row, column, physical line and byte offset, everything that would change on a parse-serialize-parse round trip or when opened in Excel or Sheets: leading zeros stripped, integers beyond 2^53 or 15 significant digits, scientific-notation and date coercion ("1E5", "1-2", "MARCH1"), formula injection (= +

Does this date fall on the claimed weekday? (batch, year boundaries and leap days) — Checks up to 200 claims 'Monday 2026-12-28' against the proleptic Gregorian calendar and reports the real weekday, whether the date exists (2026-02-29 does not), and the nearest date carrying the claimed weekday; a claim 'Friday, 13 March' without a year returns the years in a range where it holds. Built for an agent verifying dates in a document, email or itinerary. Gregorian only (no Julian date

Can this plan of steps finish inside the remaining time, in the expected case and the worst case? — Takes steps with expected and worst-case durations, dependencies (steps with no dependency between them run in parallel), retries with backoff, and the time remaining, and computes the critical path for both cases. Verdict guaranteed (worst case fits: pass), likely (only the expected case fits: unknown) or infeasible (even the expected case does not fit: fail), with the slack and the critical-path

Group webhook deliveries by identity within a time window and decide: process, dedupe, retry, reconcile or abstain — Given previously seen deliveries (identity, receivedAt, payload or digest, handling outcome) and a new incoming one, finds earlier deliveries with the same identity inside the window and returns process (none), dedupe (same payload already processed), retry (earlier ones all failed), reconcile (an earlier one is in flight or unknown, or the same identity arrives with a different payload) or abstai

Find unresolved references (that one, the former, tomorrow, next week) and return clarification questions — Scans a user or tool message for phrases that cannot be understood without context: demonstratives ('that one', 'the latter'), back-references ('as before', 'the usual'), and relative dates or times ('tomorrow', 'next Friday', 'end of month', 'soon'), each with offsets and a clarification question. Given a caller anchor (asOf instant plus IANA time zone) it resolves only unambiguous day offsets (t

Check that a sub-agent delegation states objective, acceptance criteria, budget, evidence and stop conditions — Checks the brief an agent is about to hand to another agent: the required parts are present and non-empty (objective, acceptance criteria, a budget with at least one positive numeric bound, required evidence, stop conditions; output format if required), nothing is duplicated, and acceptance criteria are not made only of vague wording (a small labelled word list: good, appropriate, if possible ...)

Dependency Underuse & Replacement Advisor — Audits the dependencies of a public GitHub repo, commit or pull request (or uploaded manifests and sources) and returns a pass/warn/fail verdict, a 0-100 risk score, the top findings with manifest file:line and the fix, and a SARIF 2.1.0 log for GitHub code scanning. Parses JS/TS, Python and Rust imports with a tokenizer-level scanner and maps them to declared npm, PyPI and crates.io dependencies.

Developer pipeline toolkit: 9 deterministic CI/CD checks — One endpoint, nine deterministic CI/CD primitives, each a sub-millisecond algorithmic check with cited specs: jwt-verify (RFC 7515/7518/8725), cron-explain (POSIX + Quartz), openapi-breaking-change (OAS 3.1 diff), webhook-verify (Stripe/GitHub/Shopify/Slack/Twilio/Linear/Clerk/SendGrid), csp-audit (CSP Level 3), cors-preflight (Fetch spec §3.2/§4.10), x509-inspect (RFC 5280/6125), graphql-cost (Gr

Verifiable dice roll from a seed, with a re-derivable transcript — Rolls dice notation (for example 4d6kh3+2) from a caller seed with a documented SHA-256 counter-mode stream and returns every die, the kept/dropped marks and a full draw transcript, for agents and referees that need a roll anyone can re-derive. Optionally checks a claimed total or claimed faces against the re-derived roll. Does NOT provide unpredictable randomness: the seed is public input, so any

PR / Dependency Breaking-Change & Exploit Gatekeeper — Merge gate for a pull request: pass/warn/fail verdict, 0-100 risk score, the top findings with file:line and a fix, drop-in patches, GitHub Checks annotations and a SARIF 2.1.0 log. Accepts a raw unified diff, a GitHub PR, commit or compare URL, or per-file patches. Deterministic scanners (secrets, SQL and command injection, weak crypto, risky package.json changes) run beside a model review for bu

UMA Dispute EV Desk — For a Polymarket market inside its roughly two-hour UMA challenge window, reads the live proposal on Polygon (adapter, optimistic oracle, bond, finalFee, expiry), judges P(proposal wrong or too early) against the literal rules with an ensemble plus red-team pass and verified quotes (full-game sports markets are decided by code with no model when Gamma's timestamps show the answer came before the g

Dockerfile build-cache and reproducibility lint — Parses a Dockerfile (continuations, escape directive, heredocs, flags, stages) and reports instruction orderings that defeat the layer cache or make rebuilds drift, for agents writing or reviewing container builds. Cache: COPY of the whole context before a dependency or package install, apt-get update in its own RUN, per-build ARG/ENV values (build date, git sha) declared before later RUN steps. R

Document to Markdown with per-page hashes and offsets — Converts a PDF (text layer), DOCX, XLSX or PPTX file, or normalizes an HTML, Markdown, text, CSV or JSON file, into Markdown (or plain text) for an agent that must read or cite a document. Give an https URL or up to 4 MB as base64. Returns the content plus one unit per page, sheet or slide with its exact character range in the content and a SHA-256 of that unit's text, so later quote and citation

Is this document still current, or superseded by a newer one? — Given the target document id, the supplied set of documents (id, effective date, which ids each one supersedes, withdrawn flag) and a reference date, reports whether the target is superseded or withdrawn as of that date, following supersession chains, and names the superseding document and its date. For an agent about to rely on a standard, price list, policy or datasheet revision. It can only see

DOI check: exists, retracted, or fabricated (batch of 20) — Checks up to 20 DOIs in one call against Crossref (including its Retraction Watch notices), OpenAlex and the doi.org Handle registry. Per DOI: FOUND with title and year, RETRACTED, EXPRESSION_OF_CONCERN, CORRECTED, REGISTERED_NOT_INDEXED, NOT_REGISTERED (no such DOI anywhere: likely fabricated), INVALID_DOI or UNVERIFIED. For research agents and RAG pipelines that cite papers: run every DOI throug

Check whether a saved interaction recipe still applies to the current version of a page — Compares a recipe (ordered steps naming selectors from an earlier dom-skeleton call), the saved skeleton elements those selectors came from, and the current page HTML. Per step it reports matched (same selector, same tag, type, role and name), changed (selector still exists but the element differs), moved (the same element now has a different selector, given as newSelector), ambiguous (several cur

List the interactive elements of an HTML page with stable CSS selectors — Reduces an HTML document to the elements an agent can act on: links with href, buttons, inputs (not type=hidden), selects, textareas, [role], [onclick], contenteditable, each with tag, type, role, best-effort accessible name (including label associations) and a CSS selector that matches exactly that element, preferring a stable id, then a data-testid-style attribute, name, aria-label, then a struc

DST gap and overlap check for local times and stated offsets in an IANA zone — Checks up to 100 local date-times (YYYY-MM-DDTHH:MM) or offset timestamps against an IANA time zone and reports whether each wall time exists once, never (spring-forward gap) or twice (fall-back overlap), with every real instant and the offset in force; also lists a zone's offset transitions in a year. Built for an agent scheduling a meeting, cron job or deadline in someone else's zone. Uses the r

Duplicate JSON object keys found in raw text, before a parser drops them — Scans raw JSON text (RFC 8259) for repeated object member names at any depth, including inside arrays and keys spelled differently but equal after unescaping ("a" vs "\u0061"), and reports each with its JSON pointer, UTF-8 byte offset, line, column and the value every occurrence carried, plus whether the values conflict. Also reports a leading BOM, unpaired surrogates, proto keys and the exact

Check that an edit's old-string occurs exactly once in a file, with match contexts — Counts how many times an old-string occurs in a supplied file text (as a replace-in-file tool would see it) and returns each match's offset, line, column and surrounding lines, for an agent about to issue an edit. pass: exactly one literal match, even counting overlapping ones. unknown: several matches, or one match that overlaps another occurrence of the same text (ambiguous; contexts show how to

Electrical arithmetic and resistor colour codes — Computes Ohm's law and power from any two of V, I, R, P; series and parallel resistance with optional tolerance bounds; a voltage divider (with optional load); an LED series resistor rounded up to a preferred E-series value; conductor voltage drop from a caller-supplied resistivity; decodes or encodes 3 to 6 band IEC 60062 resistor colour codes; and tests whether a value is an official E3 to E192

Map free-text values onto a closed set of allowed values, or abstain — Maps each input string onto one member of a closed caller-supplied set using, in order: exact match, caller aliases, a deterministic normal form (Unicode NFKC, case and accent folding, punctuation and underscore collapse), then bounded edit distance (optimal string alignment, so a swapped pair of letters costs 1). Returns the canonical value, the method, the distance and the nearest candidates; it

Check that every claim in an evidence bundle has enough independent, fresh, quote-verified support — Checks a bundle an agent assembled before answering: each claim must be backed by at least N evidence items from distinct source groups, each item's quote must actually appear in the supplied text (with offsets), publication dates must be within the caller's maximum age relative to a caller-supplied asOf, and duplicates are collapsed. Does not fetch URLs or judge whether a source is reliable or wh

Check that evidence covers the scope of the claim it is used for (geography, population, period, version, basis, unit) — Compares the scope descriptors of a claim and of the evidence offered for it, per dimension: geography and population (every claimed item must be equal to, or contained in a broader evidence item according to the caller's taxonomy), period (the evidence period must contain the claimed period; partial or no overlap is a mismatch), and version, basis and unit (must be equal after normalization). Des

Which GPS, serial-number and owner EXIF tags an image carries (keys only) — Lists which privacy-relevant EXIF tags exist in a JPEG, TIFF, PNG (eXIf) or WebP (EXIF) image, for an agent that is about to publish, upload or forward a photo. Reports tag ids and names for GPS, body and lens serial numbers, camera owner, artist, unique image id, device, software, timestamps and thumbnails; values are never returned. Input is the image as base64 (up to 96 KiB, enough for the meta

Photographic exposure value: solve aperture, shutter, ISO or EV, with equivalent exposures — Given any three of aperture (f-number), shutter time in seconds, ISO and scene EV at ISO 100, solves the fourth from EV = log2(N^2/t) and EV_S = EV_100 + log2(S/100), names the nearest one-third-stop f-number, and lists the equivalent shutter time at each f-number from f/1 to f/32. Given all four, checks that they agree within a tolerance in stops. Built for camera, lighting and image-generation a

Farmland Soil, Water & Drought Report — Before buying, leasing or planting a US parcel, reads the USDA NRCS soil survey at a point (lat/lon or US address): map unit, prime farmland class, land capability class and subclass, drainage, hydric rating, flooding and ponding frequency, water table and bedrock depth, slope, pH, organic matter, available water capacity (percent-weighted across components) and building-site ratings with shrink-s

SEC Filing Integrity Check: XBRL Footing, Tie-Outs, Revisions — Before a figure from a US 10-K or 10-Q is quoted, re-checks it with exact integer arithmetic on SEC XBRL data: balance sheet identity (including noncontrolling and temporary equity), net income bridge, cash flow to balance sheet cash tie-out with restricted cash and roll-forward, EPS recomputation, every calculation-linkbase relationship the filing declares, and same-period values that a later 10-

Fiscal year, quarter and period for a date: month-aligned or 52/53-week (4-4-5, 4-5-4, 5-4-4) calendars — Maps up to 100 dates to a fiscal year, quarter, period and week for a caller-defined fiscal calendar: a month-aligned year (starts on day 1 of a named month) or a 52/53-week year ending on a given weekday last in or closest to a month end, in a 4-4-5, 4-5-4 or 5-4-4 pattern with the 53rd week appended to the final period; detects 53-week years and checks an asserted fiscal year, quarter or period.

Fleet memory: your agents' shared observations of endpoints, rate limits, IVR paths and ambiguous submissions — Lets every agent in your fleet record what it observed and read what the others saw, with freshness: endpoint brownouts and latency, rate-limit behaviour, IVR menu paths that worked, whether an ambiguous submission was later accepted or rejected, fixes for an error signature, stale feeds, page recipes, merchant workflow failures, document reputation. record() appends timestamped observations for a

Floating-point stability of an expression over variable ranges — Evaluates a real-valued expression in binary64 (or binary32) on a deterministic grid over the caller's variable ranges and compares each point with a certified arbitrary-precision reference, reporting the decimal digits lost, the worst input, and overflow, underflow or NaN. Recognises and rewrites exp(x)-1, 1-cos(x), log(1+x), sqrt(a)-sqrt(b), a^2-b^2, the quadratic formula, sqrt(x^2+y^2) and log-

Font converter: TTF/OTF, WOFF and WOFF2 containers, with the font's own embedding permission checked — Repackages a font you supply as base64 (up to 2 MB) between the sfnt (.ttf/.otf), WOFF and WOFF2 containers, for an agent or build step that needs a web font in a particular format: every table is carried over unchanged, the result is read back and its table digest compared with the source, and the font's own OS/2 embedding bits are checked first (a restricted font fails and nothing is returned).

Font coverage: does this font have glyphs for this text, and does it fit the box — Reads a TrueType or OpenType font you supply as base64 (up to 2 MB) and a text of up to 5000 characters, for an agent about to render text in a fixed font: lists every code point with no glyph in the cmap (formats 4 and 12) and sums the hmtx advance widths to test a width in pixels. Every offset is bounds-checked. It does not apply kerning, ligatures or shaping (combining marks and complex scripts

@font-face block generated from the font file: family, weight, style, stretch and unicode-range read from the font — Reads a font you supply as base64 (up to 2 MB, any of TTF, OTF, WOFF, WOFF2) and writes an @font-face rule for the URLs you will serve it from, for an agent or build step generating CSS: font-family from the name table, font-weight from OS/2 (or the fvar wght range of a variable font), font-style, font-stretch, and unicode-range from the cmap. Names and URLs are escaped as CSS strings. It does not

Font to SVG: a line of text drawn as SVG paths from the font's own TrueType outlines — Reads a TrueType-outline font you supply as base64 (up to 2 MB; TTF, WOFF or WOFF2) and returns up to 200 characters of single-line text as an SVG made of outlines (one per distinct glyph, placed with ), plus each glyph's path data in font units (y up), for an agent producing a logo, icon label or image that must not depend on the font being installed. Composite glyphs and their transf

Attach footnote texts to their in-text markers from page layout coordinates — Given footnote markers (label, page, x, y) and footnote texts (label, page, x, y) extracted from a laid-out document, pairs each marker with its note by label and page, pairing repeated labels on one page in reading order, optionally letting a note continue on the next page, and reports unattached markers, orphan notes, ambiguous groups (different counts for the same label) and notes placed above

Cross-Lingual Dark Science Arbitrage Scout — Finds scientific work that exists only in non-English literature. Give an English topic, a technical bottleneck, or a paper you know (seedPaper: DOI or arXiv id); the scout plans native-terminology searches in up to 6 languages (default zh, ja, de, fr, ko) across OpenAlex, Crossref, HAL, J-STAGE and Europe PMC, drops works that already have an English twin, and returns up to 8 ranked discoveries.

Government Form & Procedure Finder (California, San Diego) — Given a plain-language situation (for example a city took belongings in an encampment sweep, a parking ticket, a contractor overcharge), returns the official government or court forms and procedures that apply, each with its official URL, issuer, how and where to file, fee, and the filing deadline computed from a reference date where a rule sets one. Covers California statewide plus the City and C

Formula weight from a chemical formula (Hill notation, IUPAC/CIAAW weights) — Parses a chemical formula (parentheses, brackets, hydrate dots such as CuSO4·5H2O, an ignored charge) for a chemist or an agent that must check a molar mass, and returns the Hill-notation formula, the formula weight in g/mol from the CIAAW abridged standard atomic weights summed exactly in decimal arithmetic, and mass fractions. Pass claimedWeight and tolerance to check a stated molar mass. Elemen

Open Research Opportunity Miner — Leads with a verdict (OPEN_GAPS_FOUND, NO_CLEAR_OPEN_GAPS or NO_GAPS_EXTRACTED), a confidence, a headline naming the top gap with its quoted evidence, and the next step. Collects papers for a topic or seed list (arXiv ids or URLs, DOIs, PMIDs; arXiv and OpenAlex search), reads full text from ar5iv or Europe PMC, and extracts future-work, limitation and open-question sentences in code with exact se

Coordinate axis-order and unit check from a CRS and its area of use — Checks up to 1000 coordinate pairs (or one bounding box) against the EPSG coordinate system you name, for agents handling GIS or map data: it flags longitude/latitude swaps, degrees supplied where metres are expected (or the reverse), values outside the CRS's area of use, and inverted bounds, with the point index and values. It does not reproject, geocode, validate geometry topology or datum, or j

Great-circle and WGS-84 ellipsoidal distance between coordinates — Computes the haversine (IUGG mean-radius sphere) and Vincenty inverse (WGS-84 ellipsoid) distance in metres plus initial and final bearings for up to 100 coordinate pairs, and optionally checks a claimed distance within a tolerance. Vincenty is millimetre-accurate except for nearly antipodal pairs, where it does not converge: those pairs return vincentyMeters null, status unknown and the haversine

GraphQL depth and static complexity score — Parses a GraphQL document (query, mutation or subscription) per the June 2018 spec and returns depth, field count, alias count, fragment list, cycle detection and a static complexity score. Complexity uses the common 'Σ field-score × list-multiplier' model: a listFields config maps parent-type + field-name to either a constant multiplier or a field argument name to read (e.g. 'first', 'limit'), wi

GTIN hierarchy: GTIN-8/12/13/14 forms, GS1 prefix, packaging-level link — Takes up to 100 GTINs and, for an agent reconciling a product feed or order lines, validates each check digit, converts it to the 14-digit form, reads the indicator digit and GS1 Prefix (restricted-circulation, ISBN/ISSN range, coupon), and, when a parent GTIN is given, checks the two share one base item (the packaging-level link). It does not look up the brand owner, the product or the country of

Signed hash receipt (Merkle batch, RFC 9162) — Timestamped, signed receipt for up to 256 SHA-256 hashes in one call: an RFC 9162 Merkle root, a per-hash inclusion proof, and the hub's W3C VC + EIP-191 signatures over both. For agents that need third-party evidence of what they did and when (action logs, generated files, prompts, tool outputs) without sending the data itself. Pure code, no model; verify free at /api/v1/verify-provenance.

Self-Healing Scraper & DOM-to-JSON Compiler — Synthesizes durable CSS selectors (Cheerio; no XPath) for a target JSON schema, proves them against the live DOM in-request, and returns the extracted data plus a ready-to-run TypeScript Cheerio extractor.

Extract hedging and certainty cues (may, probably, clearly, approximately) from text with offsets — Finds hedge and booster cues in an agent's or source's text using a fixed English lexicon (modals, epistemic verbs, adverbs and adjectives, approximators, shield phrases, certainty boosters) plus caller-supplied phrases, and returns each cue with its category, the sentence it sits in, and offsets in UTF-16, code points, UTF-8 bytes and graphemes. Rule-based and English only: a cue is a word match,

HTTP cacheability, CORS and redirect-loop explainer — Explains from supplied response headers (and optionally the request's headers, a CORS request, and a redirect chain) whether a shared and a private HTTP cache may store the response (RFC 9111 section 3), its freshness lifetime and basis (s-maxage, max-age, Expires minus Date, or the 10% Last-Modified heuristic), Vary, invalid max-age or Expires, the CORS check and preflight outcome per the Fetch S

Classify an HTTP response as success, retry, reconcile or stop, with the wait it asks for — Classifies one HTTP response (method, status, headers) as success / retry / reconcile / stop for an agent that just called an API, and reads the wait from Retry-After (RFC 9110: delay-seconds or any of the three HTTP-date formats) or from a vendor delay header whose dialect the caller declares (seconds, milliseconds, epoch seconds, epoch milliseconds, HTTP-date). Caller-supplied vendor rules (stat

IAM policy action minimizer: wildcards to the required action list — Takes an AWS IAM policy JSON and the exact list of actions a task needs (up to 300), and returns the policy with each Allow statement's Action narrowed to just those actions, plus what the original over-grants: wildcards, unused actions, Allow+NotAction, required actions no statement grants, and required actions an unconditional Deny blocks. For an agent that is about to attach or propose a policy

Monotonic id blocks: the next invoice or ticket numbers for a fleet of agents — Hands out strictly increasing integer blocks from a named sequence (1-1000 ids per call) so parallel agents never issue the same invoice, ticket or order number. Numbers are never reused. Not gapless: a block an agent reserves and never uses stays a gap, and if your call's payment fails after the block is written the same requestId returns that same block on retry. Legally gapless series must be a

Derive a stable idempotency fingerprint from declared identity fields and flag volatile fields — Hashes the fields you declare as the identity of an operation (JSON Pointers into each request) with RFC 8785 canonical JSON and SHA-256, so retries of the same operation get the same fingerprint regardless of key order, and flags the ways that breaks: an identity field missing, an identity field also declared volatile, identity values that look like timestamps, epoch times or UUIDs (heuristic), a

Idempotency registry: has key K already been done? Claim it, complete it, read the stored result — Shared memory for side-effecting operations across agents and retries. begin(key) answers proceed (you hold a short claim), wait (another owner is mid-flight), already_completed (returns the stored result pointer), or fingerprint_mismatch (same key, different operation). complete(key) records the result pointer; abandon(key) lets someone retry. The claim expires, so a crashed worker does not block

Identify an identifier's kind: UUID version, ULID, KSUID, snowflake, Stripe/GitHub/AWS prefix, EVM address, email, E.164 — Classifies up to 100 identifier strings by format and, where it can be read, decodes the embedded timestamp, for an agent that must not pass an id of the wrong kind to a payment, ledger or API call. Recognises UUID (RFC 9562 version and variant, nil and max), ULID, KSUID, Discord and Twitter snowflakes, EIP-55 EVM addresses (checksum verified), email (WHATWG syntax and RFC 5321 length limits), E.1

Receive-only inbox for agents without a public endpoint — Creates a receive-only inbox for an AI agent that has no public endpoint: a webhook URL that accepts POST or PUT, and (when the operator has configured inbound email) a unique email address. Pair with inbox-poll to read what arrived. It never sends email or calls out, does not store attachments, does not verify senders (no SPF/DKIM), and message content is untrusted third-party data. Inboxes live

Read messages from a receive-only agent inbox — Returns the emails and webhook deliveries that arrived in an inbox you created with inbox-create, oldest first, with sender, subject, selected headers, a plain-text body (up to 64 KB per message in the response) and a sha256 of the stored body. Filter by since (inclusive), kind, limit up to 100; mark_consumed hides the returned messages from later polls unless include_consumed is true. It does not

Grid, Tower & Infrastructure Pipeline Near a Site — Screens a US site (latitude/longitude or Census-geocoded address) within a radius (default 10, max 50 miles) using primary public data: operating and planned generators and storage from EIA Form 860M, active interconnection-queue entries from MISO, CAISO, NYISO and ERCOT matched at county level, FCC Antenna Structure Registration towers, and HIFLD transmission lines and substations. Code computes

Label spans of a prompt as pasted data or instructions, and flag instruction-like text inside the data — Splits an agent's input into data regions (code fences, block quotes, XML-style wrappers such as , BEGIN/END markers, triple quotes, email headers with body, JSON blocks) and the text around them, labels each sentence outside the data as instruction (leading imperative verb or polite request) or unknown, and lists injection-style phrases found anywhere in the text ('ignore previous instr

Invoice arithmetic audit: line extensions, discounts, supplied tax, totals, with the drift in minor units — Recomputes up to 500 invoice lines in exact decimal arithmetic and compares them with the stated amounts, for an agent about to approve, pay or reconcile an invoice. Checks: quantity x unit price less discount equals the line amount; tax is taken as supplied (tax amounts are summed, never looked up) and, where a rate is given, tax equals rate x line amount under the declared rounding scope (per li

Row multiplication, unmatched keys and uniqueness violations before a table join — Takes the join keys of two tables (every row's key, or distinct keys with row counts) and computes the exact output row count of an inner, left, right or full equi-join with SQL NULL semantics, the observed cardinality (one-to-one, one-to-many, many-to-one, many-to-many), the keys that multiply rows the most, unmatched keys on each side and the first duplicate row when a side was assumed unique. I

Conflicts between two to four RFC 6902 JSON Patches made against the same document — Takes one base document and 2 to 4 JSON Patch (RFC 6902) operation lists written against it and reports, per pair, whether they commute: same-path writes, ancestor/descendant overlaps, array-index shifts (an insert or remove moves the elements a later patch addressed by index), concurrent - appends whose final order depends on application order, test operations invalidated by another patch, an

JWT lint: alg none, crit, key-URL headers, expiry, audience (decode only) — Decodes a compact JWT (never verifies the signature) and checks it against RFC 7519 and RFC 8725 for an agent about to trust or forward it: alg none or missing, crit and key-supplying headers (jku, x5u, jwk), kid injection shape, duplicate claims, and exp, nbf, iat against a caller-supplied time, plus optional expected algorithms, issuer and audience. Without asOf the time claims are not judged an

JWT signature and claims verification — Verifies a JWS-compact JWT: signature over base64url(header).base64url(payload) with Web Crypto under the exact algorithm in the token's alg header (HS256/384/512, RS256/384/512, PS256/384/512, ES256/384/512, EdDSA). Refuses 'none' unconditionally and rejects any alg not in the caller's allowedAlgorithms. Validates exp/nbf with a configurable clock-skew tolerance (default 60s), iss equality, a

Kubernetes manifest cross-reference check — Parses up to 20 Kubernetes YAML manifests and checks that they agree with each other, for an agent that generated or edited manifests before kubectl apply: Service selectors against pod template labels (with the nearest near-miss), named targetPorts against containerPort names, Service port naming, Deployment selector against its template labels, Ingress backends against Services, and ConfigMap/Se

Autonomous Lead Enrichment — Scrapes a company homepage, deduces its value proposition and buyer persona, and drafts a targeted B2B cold email.

Short lease with fencing tokens: claim a task for N seconds, renew, release — Lets many agents agree which one works on a task: acquire a named resource for 1-3600 s, renew it while working, release it when done. Every new acquisition returns a strictly larger fencing token; pass it to the downstream store and reject writes carrying a smaller one, so a paused holder whose lease expired cannot clobber its successor. Re-acquiring as the current holder extends the lease with t

Legal citation and quotation verifier (US case law) — Extracts US reporter citations from a brief (U.S., S. Ct., L. Ed., F., F.2d, F.3d, F.4th, F. Supp., state and regional reporters) and checks each against Harvard's Caselaw Access Project and, when the operator has configured a token, the CourtListener Citation Lookup API. Reports FOUND, MISMATCH (case name, year or circuit differs), NOT_FOUND, AMBIGUOUS, or UNCHECKED with the reason, and never mar

SPDX license identification from LICENSE text — Normalizes a LICENSE file (SPDX varietal substitutions: whitespace, case, quotes, bullets, copyright lines) and matches it against distinctive phrases of 32 common SPDX licenses (MIT, MIT-0, Apache-2.0, BSD-2/3/4-Clause, 0BSD, GPL-2/3 +/- or-later, LGPL-2.1/3.0 +/- or-later, AGPL-3.0 +/- or-later, MPL-2.0, ISC, Unlicense, CC0-1.0, Zlib, Artistic-2.0, EPL-2.0, BSL-1.0, WTFPL, Python-2.0, PostgreSQL

Literary Agent Scout: quoted wish-list matches with query status and fee checks — For an author with a book to query, finds literary agents and agencies on the agencies' own websites, reads their guidelines and agent bios, and ranks them against your pitch. Every fit claim carries a sentence quoted verbatim from the agent's page (quotes not found on the page are dropped), each agent shows whether the page says queries are open or closed with the retrieval time, and any agency p

Detect LLM text watermark evidence: green-list z-test for a key you hold, plus invisible-character markers — Tests whether text carries watermark evidence, for a pipeline that must decide whether to trust or label content. Two independent checks: (1) a green-list z-test (Kirchenbauer et al. 2023) over green flags you supply, or over token ids and a key using this endpoint's own HMAC-SHA256 scheme, returning the z-score and exact binomial p-value; (2) an invisible-character scan of the text for zero-width

Parse locale-formatted money text (1.234,56, 1,00,000, accounting parentheses) without guessing — Parses up to 100 number or money strings such as "1.234,56", "1,00,000", "(1,234.50)", "￥1,000" or "CHF 1'234.50" into canonical decimal strings, for an agent reading amounts from documents, emails or tables. With a BCP 47 locale it applies that locale's CLDR separators (read from Intl.NumberFormat); without one it lists every valid reading and an ambiguous string such as "1,234" returns unknown w

Resolve git conflict markers in package-lock.json or a YAML lockfile — Takes a lockfile (package-lock.json or a YAML lockfile such as pnpm-lock.yaml) containing git merge-conflict markers, rebuilds both sides, and merges them key by key (using the diff3 base when present), returning the resolved file or the exact conflicting key paths with both values, for agents rebasing or merging branches. An optional prefer=ours|theirs applies one explicit policy to the remaining

Collapse a log into error signatures with counts and first/last occurrence — Groups the records of a log or CI output by signature: timestamps, UUIDs, IPs, hex ids, quoted values and numbers are replaced by placeholders, stack-trace continuation lines are attached to their record and contribute their first frames, and each signature reports its count, severity, first and last line, first and last timestamp (as written, when the line starts with an ISO 8601 timestamp) and a

Does this value survive conversion to another numeric or spreadsheet type? Range, precision, leading zeros, E notation — Takes raw JSON literals (as written, so 9007199254740993, 1e400, -0, "007" and 0.10 are judged on their digits, not on a parsed double) and target types (binary64, binary32, int8..int64, uint8..uint64, PostgreSQL numeric(p,s), Arrow decimal128/decimal256, Parquet decimal, Excel cell, pandas int64) and reports for each pair whether the value is exact, lossy (rounded, truncated, leading zeros or sig

Choose which inventory lots to pick (FIFO, FEFO, LIFO, fewest lots) — Allocates a demand quantity across up to 200 inventory lots under a stated policy (FIFO by received date, FEFO by expiry, LIFO, or fewest lots), excluding held, expired, short-shelf-life and wrong-location lots, in whole case multiples if asked. For warehouse and order agents. Returns the pick plan, the shortfall and a reason for every excluded lot. A caller asOf date is required (the check never

File type vs leading bytes (magic numbers, disguised HTML, truncation) — Compares a claimed file type (MIME type and/or file name) with the first bytes of the file, for an agent that is about to open, parse or forward a download. Detects PNG, JPEG, GIF, WebP, ZIP and OOXML (docx/xlsx/pptx), GZIP, PDF, MP4 (ftyp), ELF, PE, RAR, BMP, ICO and HTML, flags an HTML login or error page saved as a PDF/image/archive, and reports truncation from end-of-file markers when the last

Does this manual or procedure apply to this unit? — Checks whether a manual, service bulletin or procedure applies to a specific unit by comparing the scope the document declares (models, firmware or revision range, serial range, exclusions) with the unit's facts the caller supplies. Versions written as dotted digits (2.10.1) compare numerically per segment. For an agent about to follow a repair or configuration procedure. Applicable only when ever

Manuscript Pre-submission Check — Verifies every reference in a manuscript against Crossref, OpenAlex, arXiv and Europe PMC (fabricated, mis-cited, retracted) and runs deterministic prose forensics on the same text. Runs reference-verifier ($0.50) + audit-prose-forensics ($1.00) in one paid call for 20% less than buying them separately. Charged only when every member succeeds.

Normalize sloppy model-generated Markdown to one house style, leaving code untouched — Rewrites Markdown from a language model into a consistent style and reports each edit by line: restores the space in '##Title', converts '*', '+' and '•' bullets to your chosen marker, strips trailing whitespace (keeping two-space hard breaks), collapses runs of blank lines, puts blank lines around headings and fenced blocks, closes a fence left open at the end, and ends the file with one newline.

Prediction Market Analyzer — Independent probability that a Polymarket or Kalshi market resolves Yes, checked against the live book. Pass a market (slug, URL, conditionId or Kalshi ticker) and/or news text: the agent reads the rules, end date and best asks, estimates 0-100 with live web search (falling back to model knowledge if search is slow), and returns verdict.action BUY_YES / BUY_NO / NO_TRADE with edge in bps after tak

Is the exchange in its core session at this instant? (NYSE holidays and 13:00 early closes, or a caller calendar) — For up to 50 caller-supplied instants, reports whether the NYSE core session (09:30-16:00 America/New_York, 13:00 on early-close days) is open, the reason when closed (weekend, named holiday, before open, after close), that day's open and close instants and the next open; or the same for a caller venue (zone, open, close, holidays, early closes). NYSE data is a dated snapshot of nyse.com covering

MCP tool pin: detect changed tool definitions (rug pulls) and hidden Unicode — Send an MCP server's tools/list result (up to 200 tools) and, after the first call, the hashes you approved. Returns a reproducible manifest hash, a per-tool hash, which tools were added, removed or changed since approval (silent description and schema edits are how MCP rug pulls and tool poisoning land), and Unicode smuggling findings in every name, description and schema string. Call it at sessi

Medical Bill Upcoding & Overcharge Slasher — Paste an itemized hospital or physician bill and get a decision first: dispute it or not, the estimated dollars above a defensible price, and a dispute letter ready to send. Flags E/M upcoding, NCCI unbundling, duplicates and unit errors line by line, prices professional claims against the CMS physician fee schedule, checks the No Surprises Act self-pay dispute deadline, and lists the letter's pla

STL mesh check: open edges, non-manifold edges, inverted normals — Checks a binary or ASCII STL mesh of up to 4000 triangles for a 3D-printing or CAD agent: open (boundary) edges, edges shared by more than two triangles, inconsistent winding, degenerate and duplicate triangles, normals that disagree with the vertex order, and an inside-out (negative signed volume) closed mesh, with triangle indices as witnesses. Input is base64 of the STL file (up to 200084 bytes

METAR / SPECI decoder with flight category — Decodes up to 20 METAR or SPECI aviation weather reports (US and ICAO forms: wind with gusts and variable direction, visibility in statute miles or metres, CAVOK, RVR, present weather, cloud layers, temperature/dew point, altimeter, trend, RMK remarks) into structured fields and computes the FAA flight category (LIFR/IFR/MVFR/VFR). Every token it cannot decode is returned with its character offset

Tiered usage pricing: graduated or volume tiers, included quantity, package units, minimum charge — Prices a usage quantity against up to 50 tiers, graduated (each tier prices only its own units) or volume (the tier holding the whole quantity prices all of it), for an agent checking a usage bill or quoting a plan. Order of operations: subtract the included quantity, convert to billing packages (per N units, rounded up, down or not at all), price the tiers, add flat fees, apply the minimum charge

Whether two metric definitions measure the same thing, and which fields differ — Compares two structured metric definitions field by field (numerator, denominator, aggregation, weighting, population, filters, exclusions, period grain and calendar, timezone, currency, FX and tax basis, unit) and returns which differences make the numbers not comparable, which are adjustable with raw data, and which fields are missing from one side. In drift mode a is the old and b the new defin

Run up to 20 micro checks under one payment ($0.02) — Bundles up to 20 calls to any pure micro check (catalog entries that are neither stateful nor network-reading, e.g. checksum, patch-anchor, allocate, business-deadline) into one x402 settlement, so a full batch costs $0.001 per check. Every call is validated before payment; one invalid call rejects the batch with its index. The request is capped at 128kb and a CPU weight of 40 (an ordinary check w

Migration DDL: data-loss and rollback hazards per dialect — Classifies each statement of a PostgreSQL, MySQL or SQLite migration (up to 20000 characters) as reversible, needing a backup, needing the original definition, or irreversible, for agents that write or review schema migrations. Flags DROP TABLE/COLUMN, TRUNCATE, unbounded DELETE/UPDATE, lossy type changes, NOT NULL or constraints that fail on existing rows, table-rewrite and lock hazards, and MySQ

Grant Milestone Verifier with Signed Release Attestation — Decides whether a grantee met a milestone before a DAO releases the next tranche or stream. Checks GitHub as of the due date (commits, tags, releases, CI, files), npm/PyPI/crates.io publish times, Snapshot proposal state, on-chain deployments (bytecode selector scan through proxies, Sourcify, creation transaction), and URLs; judges the rest with an ensemble plus a skeptical reviewer, reads Sablier

ISO 4217 minor units: decimal amount to integer minor units and back — Converts up to 200 money amounts between decimal strings and integer minor units (cents, yen, fils) using the ISO 4217 exponent from the SIX Group list, for an agent about to send an amount to a payment API. Returns the exponent, the integer minor amount and the canonical decimal string; JPY is 0, KWD and BHD are 3, CLF is 4. An amount with more fraction digits than the currency allows fails with

Diff a before/after JSON write for deletes, null-vs-omit and emptied arrays against protected paths — Compares the JSON document an agent read (before) with the one it is about to write (after) and lists every change by JSON Pointer: removed members, null turned into omitted (and the reverse), arrays emptied or shrunk, objects emptied, type changes. Fails when a change touches a protected path or is destructive and not listed in allowDestructive; use it before a PUT, PATCH or bulk update. Arrays a

Which timestamps are anchored? Flags naive local times, zone abbreviations, date-only, epoch units and d/m ambiguity — Classifies up to 200 timestamp strings as anchored (RFC 3339 with Z or offset), naive local (no offset), date-only, epoch seconds or milliseconds, zone abbreviation, ambiguous numeric date, or unparseable, and lists the hazard of each; with assumedTimeZone it also reports whether a naive time falls in a DST gap or overlap there, every instant it could mean, and offsets that contradict the zone. Bu

Find negation cues and the words they cover; report whether given terms fall inside a negation scope — Applies NegEx-style rules (the English termset of negspacy, MIT, derived from NegEx) to text: finds negation cues, pseudo-negations such as 'not only', and scope terminators such as 'but', and returns each cue's scope window with offsets in UTF-16, code points, UTF-8 bytes and graphemes. Given target terms, reports per occurrence whether it is negated, not negated by the rules, or unknown (near a

Number and cheque-amount words: English (US, UK, Indian lakh/crore) — Spells out a number or a money amount in English words for a cheque, contract or invoice, for an agent that must write the amount in words next to the figures. Locales: en-US ("one hundred twenty-three"), en-GB ("one hundred and twenty-three"), en-IN (lakh and crore, "one crore twenty lakh"). With a currency it returns the integer words, the minor-unit fraction as digits over 10^exponent ("56/100"

OpenAPI 3.x breaking-change diff — Diffs two OpenAPI 3.x documents and classifies every delta as breaking, compatible or informational. Breaking: removed path/operation/response status/media type, type/format/pattern tightening, enum value removed, required property added, request body became required, maximum lowered or minimum raised, nullable removed. Compatible: additions (new paths, optional params, enum values, response statu

Slice one operation out of an OpenAPI 3.0 or 3.1 document with every component it needs — Extracts one operation (path plus HTTP method, matching templated paths like /pets/{id}) from an OpenAPI 3.0.x or 3.1.x document, given as JSON, YAML or a parsed object, and returns a smaller valid OpenAPI document holding that operation, its path-level parameters and servers, and the transitive closure of every local $ref it uses: schemas, parameters, requestBodies, responses, headers, examples,

Open Badges 2.0 / 3.0 credential verifier — Verifies an Open Badges credential end to end and says what could and could not be confirmed. Accepts an Open Badges 3.0 credential (JSON with an eddsa-jcs-2022 proof, or a VC-JWT signed with EdDSA, ES256 or RS256), an Open Badges 2.0 assertion (hosted, or signed as an RS256 JWS), a hosted-assertion URL, or a baked PNG or SVG image. Checks the 2.0/3.0 required fields, the issuance and expiry windo

Plot a prose intensity curve per sentence and flag where a tense passage loses velocity — Scores each sentence of narrative prose for pace from sentence length, share of one-syllable words, exclamation and question marks and, if you supply a word list, action-word share; smooths the scores into an intensity curve; and reports stalls, the spans where intensity falls well below an earlier peak. For a writing tool or editor agent checking an action or tension scene. A fixed-weight heurist

Package exists on npm, PyPI or crates.io (hallucinated dependency check) — Checks whether up to 10 package names (optionally with a version) exist in the live npm, PyPI or crates.io registry, for coding agents about to add a dependency they may have invented. Per package: exists, latest version, deprecated or yanked flag, and for a missing name the closest names the registry's own search returns (npm, crates.io; PyPI has no search API). Reads registry metadata only: neve

Pack sizes, MOQ, order increments, overbuy cap and price breakpoints: what to actually order — Finds the cheapest valid order quantity for a requested number of units across up to 50 supplier options, for an agent buying parts or consumables sold in packs. Each option has a pack size (units per pack or notation such as "6x100", "case of 12", "12/case", "case of 6 inners"), an optional minimum order quantity, an order increment and price breakpoints; the order is rounded up to a multiple of

Page capture: screenshot or PDF of a public URL — Renders one public web page in a real browser (Cloudflare Browser Run, JavaScript enabled) and returns a PNG or JPEG screenshot or a PDF as base64 with its sha256, for agents that need to see or archive a page. Does not log in, click or fill forms, does not capture private or internal hosts, and returns an error rather than a capture larger than 6 MB.

Page contents: URL to clean Markdown or text — Fetches one public web page and returns its main content as clean GitHub-flavored Markdown or plain text, with title, canonical URL, language, absolute links and content hashes, for agents that feed pages to a model. Deterministic, no model. Does not run JavaScript (client-rendered pages come back sparse), does not follow links, does not consult robots.txt, and drops navigation, footers, sidebars

Detect repeated cursors, duplicate-only pages and no progress in a pagination scan — Takes the pages an agent has fetched so far (cursor used, next cursor returned, item ids) and finds the failures that make a scan loop or lie: a next cursor already used, a page whose items were all seen before, a page that returned nothing new yet offers another cursor, a scan that has not ended. Optionally takes declared ordering and cursor assumptions and lists the omission risks they imply (of

Paper Contradiction Finder — Takes 2 to 4 papers (DOI, arXiv id, PMID or their URLs) on a shared question and answers whether their reported numbers conflict: a verdict (CONTRADICTION_FOUND / NO_CONTRADICTION_FOUND / INCONCLUSIVE) with confidence, a one-line headline, the decisive quotes side by side and a next step, then the full evidence. Fetches the best available text (ar5iv, Europe PMC full text, else abstract), extracts

Apply a unified-diff hunk or search/replace edit to a file by content, refusing ambiguity — Applies unified-diff hunks (wrong line numbers are fine) or search/replace blocks to a supplied file text and returns the patched text, for an agent whose edit tool failed or whose diff has drifted line numbers. Each hunk is located by its content against the original file, trying exact, trailing-whitespace, outer-whitespace then collapsed-whitespace matching; the rung used is reported per hunk. I

Patent Design-Around & Freedom-to-Operate Architect — Paste patent claims or give a patent number. Splits the independent claims into lettered limitations in code, charts your product against each one (literal and doctrine of equivalents), scores claims under the all-elements rule, and proposes 3 engineering design-arounds that each remove a limitation of every independent claim, checked in code. Leads with a decision and optional dollar exposure. A

Patent Strategy Pack — Both sides of a blocking patent in one call: pre-priority-date prior art charted against every claim element, and an all-elements freedom-to-operate map with three engineering design-arounds. Runs prior-art-invalidator ($1.00) + patent-design-around ($2.00) in one paid call for 20% less than buying them separately. Charged only when every member succeeds.

File path portability check across Windows, macOS and Linux — Checks up to 500 relative file paths for names that break or collide on Windows, macOS or Linux: Windows reserved device names (CON.txt, NUL.tar.gz, COM¹), reserved characters, trailing dots and spaces, component and total length limits, and distinct paths that become one file through case folding or NFC/NFD normalization. For an agent that writes files from an archive, a generated tree or a repo

Which gitignore, CODEOWNERS or glob rule matches a path — Evaluates up to 25 file paths against the text of a .gitignore, a CODEOWNERS file or a list of glob patterns and reports, per path, whether it matched, the winning rule's line and pattern, and why (last match wins, directory-only rules, negation, a re-include blocked by an excluded parent directory, CODEOWNERS owner lists). Optional per-path expectations turn it into a pass/fail check, and CODEOWN

Privilege expansion between two manifests: npm, Chrome extension, Android — Compares an old and a new manifest and reports every way the new one gains privilege, for agents reviewing dependency bumps and extension or app updates. npm package.json: new or changed install-time lifecycle scripts, gypfile, bin entries and URL/git/file dependencies (including overrides). Chrome manifest.json: added permissions with Chrome's own warning text, host access not covered by the old

Effective permissions of a delegated agent: requested ∩ user ∩ parent, minus explicit denials — Computes what a sub-agent may actually do: the permissions it requested, intersected with what the end user holds and what its parent agent holds, minus any explicit deny list, and says for each denied permission which party lacks it. Permission names are compared exactly (no wildcards); caller-declared implications (for example 'admin' grants 'read') can be given and are expanded transitively wit

Does this text fit: SMS segments, X posts, store listings, ad and push limits — Checks up to 50 texts per call against a platform limit, for an agent writing a message, post, store listing, ad or push payload that must not be truncated or split. sms: GSM 7 bit vs UCS-2 and the exact segment boundaries per 3GPP TS 23.038 (extension characters cost 2 septets and never straddle a segment; an emoji is 2 UTF-16 units). x: weighted length per the open-source twitter-text config v3

Proof-of-delivery sufficiency against a required evidence list — Checks whether the proof-of-delivery evidence supplied (signature, photo, timestamp, GPS fix, recipient name, scan ...) satisfies the carrier or contract requirement list the caller gives: every required item, plus optional 'at least k of these' groups, with value bounds (for example GPS accuracy at most 30 m) and maximum age. For an agent assembling a delivery dispute or a payment release. Return

Policy gate with a signed receipt: your JsonLogic rules vs a proposed tool call or payment — Evaluates your own policy (JsonLogic rules, each allow or deny, deny-overrides or first-match, default deny) against a proposed tool call, payment or action, and returns allow or deny, the rule id that decided, and a receipt signed by the hub over the hashes of the policy, the request and the decision. The receipt is the product: another agent or an auditor can verify offline which rule allowed th

Case-law precedent search (CourtListener, US) — Finds US court opinions most on point for a legal question or fact pattern: searches CourtListener's case-law index (optionally limited to courts and filing dates), reads the top opinions, and returns the best matches with a holding passage verified verbatim against the opinion text, plus citation, court, date, precedential status and a source hash. A second skeptical pass flags matches that look

Pre-merge Review Pack — One unified diff in, one merge decision out: the PR exploit gatekeeper (bug hunter, patch synthesizer, secret/dependency/SAST scans) plus the adversarial Unicode / Trojan Source gate. Runs diff-gatekeeper ($0.50) + unicode-safety-gate ($0.25) in one paid call for 20% less than buying them separately. Charged only when every member succeeds.

Attribute a total's change to quantity, price, FX, added and removed lines (sums exactly) — Explains why a total moved between two versions of a quote, order or invoice (up to 300 lines each), for an agent that must say how much of the change came from quantity, unit price, exchange rate, new lines and dropped lines. Lines are matched by key. Matched lines are decomposed by sequential substitution in the declared factor order (default quantity, then price, then fx), so the effects sum to

PDF page box check: trim, bleed and safe area from supplied boxes — Checks the five PDF page boxes you supply (MediaBox, CropBox, BleedBox, TrimBox, ArtBox, in points) for a print job: applies the ISO 32000-1 defaults and media-box clipping, measures bleed on each side of the trim box, and tests the content box against a safe margin inside the trim, for agents that prepare or accept print-ready PDFs. It does not read a PDF file, check fonts, colour, image resoluti

Print resolution, bleed and uncompressed bitmap size arithmetic — Given an image's pixel size and a print size (inches, millimetres or centimetres) plus optional bleed, margin and target ppi, returns the effective ppi over trim plus bleed, whether it meets the target, the pixel count a target ppi needs, the largest print size a target ppi allows, the safe area, and the exact uncompressed bitmap size. Built for print-on-demand, layout and image-prep agents; the t

Global Prior-Art Patent Invalidator — Screens a patent's claims against scholarly literature published before the priority date (arXiv, OpenAlex, Europe PMC, Crossref, HAL, J-STAGE; English, Chinese, Japanese, German; not patent databases). Give claims plus a priority date, or a patent number to read them from Google Patents. Claims are split into elements in code; every claim-chart quote is verified verbatim against the title or abst

Pronunciation lexicon to SSML phoneme and sub tags — Applies your pronunciation lexicon (up to 200 terms, each an IPA string or a spoken alias) to a text of up to 20000 characters and returns SSML 1.1 with or around each whole-word match, XML-escaped, for an agent preparing text for a text-to-speech engine. Matching is whole-word, longest term first, left to right, deterministic. It does not guess pronunciations,

Property Hazard & Site Risk Report — Before an agent recommends, underwrites or screens a US property, reads free public federal hazard data for a street address or lat/lon: Census geocode, FEMA National Flood Hazard Layer flood zone, SFHA status and base flood elevation, USGS ASCE 7-22 seismic design values, FEMA National Risk Index tract ratings (wildfire, hurricane, tornado and more), OpenFEMA county disaster declarations for 10 y

Property Code-Violation & Lien History (Open-Data Cities) — Before buying, leasing or insuring a US property, reads the city's own open-data code-enforcement records for one address (or a NYC BBL/BIN) and returns normalized violations with date opened, category, description, open/closed status, source severity class and record URL; deterministic counts by status, category and year; days since the latest violation; the longest repeat-citation streak (citati

Proration and day-count fractions (30/360, ACT/365F, ACT/360, ACT/ACT ISDA) on half-open date ranges — Computes the year fraction of up to 100 half-open date ranges [start, end) under a named ISDA day-count convention, and optionally the prorated charge, for an agent billing a partial period or accruing interest. Conventions follow Section 4.16 of the 2006 ISDA Definitions: 30/360, 30E/360, 30E/360 (ISDA), Actual/365 (Fixed), Actual/360, Actual/Actual (ISDA), 1/1. end is the day immediately follo

Requested vs quoted specification attributes: met, not met or not comparable — Compares up to 200 requested attribute requirements with the values a supplier quoted (voltage, grade, finish, tolerance, brand, certification), for an agent checking that a quote matches the request before buying. Requirements are eq, ne, gte, lte, gt, lt, between (value..max), in (a list) or present; numeric comparisons use exact decimals, an optional tolerance, and the caller's unit conversions

Would this request fit in the remaining quota, given a snapshot and pending reservations? (snapshot-bound) — Computes available quota from a caller-supplied snapshot (limit, used, reserved) minus the unexpired pending reservations and says whether a requested amount fits. This is a read of the numbers you supply as of snapshotAt: it reserves nothing, so concurrent consumers can still use the quota before you do, and every response says so. A stale snapshot gives unknown, not pass; a window that ended at

Normalise supplier quotes to one comparable unit price (currency, unit of measure, tax, discount) — Converts up to 200 quoted prices to a net price per base unit in one currency, for an agent comparing supplier quotes. Each line: price for per units of a unit of measure, optional tax-inclusive flag and rate, optional percentage discount, any currency. Uses only the caller's exchange rates and unit factors (1 uom = factor base units); a currency without a rate, a unit without a factor, or a tax

Recipe scaling with caller-declared non-linear rules — Scales up to 60 ingredient quantities by a servings ratio or a factor for an agent resizing a recipe or batch formula, applying per-ingredient rules the caller declares: linear (the labelled default), power exponent, fixed amounts, min/max clamps and rounding steps, with an optional display unit via unit-convert. Reports what rounding did (relative drift, rounds-to-zero). It does NOT know which in

Exact Ledger Reconciler — Explains which charges, refunds and fees make up a Stripe, PayPal or bank payout, with an exact solver instead of a model: subset-sum over integer minor units (meet-in-the-middle, DP, branch-and-bound) across candidate fee schedules (percentage + fixed, tiers, caps), withholding, FX rates and rounding conventions (half-up, half-even, truncate; per line or on the total). Paste a Stripe balance or p

Expand an RRULE or 5-field cron into concrete occurrences with DST handling, exclusions and conflict detection — Expands an RFC 5545 RRULE (FREQ DAILY/WEEKLY/MONTHLY/YEARLY, INTERVAL, COUNT, UNTIL, BYDAY with ordinals, BYMONTHDAY, BYMONTH, BYSETPOS, WKST; DTSTART with TZID; EXDATE) or a crontab(5) expression into up to 366 occurrences in a time zone, with the UTC instant of each, DST gap or overlap flags, caller blackout dates, a plain-English reading, and optionally the occurrences where it collides with a

Selective PII redaction with a self-check — Redacts detected personal data at the level you choose and rescans the output with the same rules. This self-check does not prove that all personal data is gone. Pick a level: 1 government identifiers (SSN, ITIN, EIN, UK NINO, Canadian SIN, driver's license, passport), 2 adds dates of birth, 3 adds street addresses and phone numbers, 4 adds person names and email addresses. Then switch single cate

Redirect chain credential-leak check — Walks a supplied HTTP redirect chain (up to 25 hops of status + Location) for an agent or HTTP client author and reports, per hop, which of the request's header names (Authorization, Cookie, Proxy-Authorization, API-key-like, Referer) are stripped or still forwarded, the method after the hop (301/302 POST to GET, 303, 307/308), whether the hop crosses origin, scheme, or registrable domain, an http

Bibliography Reference Verifier — Checks every reference in a bibliography against Crossref, OpenAlex, arXiv and Europe PMC and leads with a verdict (FIX_REQUIRED, REVIEW_RECOMMENDED, INCOMPLETE or ALL_VERIFIED), a confidence, a one-line headline, the next step and a prioritized fix list. Input: BibTeX, LaTeX \bibitem, a plain-text list (APA, IEEE, Vancouver), an arXiv id (read via ar5iv), a PDF URL, or just a list of DOIs, arXiv

Regex catastrophic-backtracking check, then test cases only if proven safe — Statically analyses an ECMAScript regular expression (up to 500 characters) for catastrophic backtracking (nested or overlapping quantifiers, ambiguous alternation in a loop, backreferences, lookarounds) for agents that write or accept regexes. Runs the caller's test subjects (up to 20, 1000 characters each) only when the pattern is proven polynomially bounded and the subject is short enough for t

Regional Economy & Trade Brief — Given a US state, county or metro (name, FIPS or CBSA), an industry (NAICS code or description) and a role (retailer, logistics, cfo, lender, other), reads BLS LAUS and QCEW, FRED (state employment, house prices, goods exports and imports, regional CPI), Census Building Permits and USAspending live, then computes levels, year-over-year and three-year changes, location quotient, z-scores against th

Regulatory Change Brief (Federal Register, EDGAR, Dockets) — Given an industry or topic and optional agencies, CFR parts, terms and companies, pulls recent US regulatory changes from official free sources: Federal Register final and proposed rules (with effective dates, comment deadlines and official URLs), SEC EDGAR 8-K items and other filings for named companies, and CourtListener dockets and opinions when the operator has a token. Days remaining are comp

Scientific Consensus & Replication Failure Auditor — Give a DOI, arXiv id, PMID, URL, title or a free-text claim and get a verdict-first screen of whether the finding has been retracted, contradicted or failed to replicate: ROBUST / CONTESTED / UNREPLICABLE_HYPE / INSUFFICIENT_EVIDENCE with a 0-100 replicationRiskScore, confidence, a DO_NOT_RELY / VERIFY_BEFORE_RELYING / RELY_WITH_CAVEATS action, a one-line headline and a next step. Checks Crossref,

Checklist coverage: caller requirements against caller evidence — Evaluates a checklist of requirements the caller defines (each made of needs on evidence keys: present, equals, one of, min/max, maximum age) against evidence the caller supplies, and returns met / unmet / unknown per requirement with the exact need and value that decided it. For an agent that must prove a task's acceptance criteria were covered before it reports done. Does not judge whether the p

Content-addressed result cache shared by your agents (short TTL) — Caches a result under the SHA-256 of the canonical (JCS) request, so every agent in your fleet that makes the same request reuses one answer for up to an hour. get returns hit with the value and age, or a miss (status unknown). Scoped to your wallet or tenantId, never shared across customers, so nobody else can poison your cache.

Decide whether a failed HTTP call may be retried, must be reconciled, or must not be repeated — Given a request's method, how it ended (response status, timeout, reset, refused), idempotency-key evidence and optional caller policy, returns retry / do-not-retry / reconcile / unknown for an agent deciding what to do after a failed tool or API call. Encodes RFC 9110 method idempotency, the proxy and retry-of-a-retry rules and Retry-After parsing; a timeout or reset is never read as success or a

robots.txt (RFC 9309) rule reading with llms.txt and AI-use signals — Reads a supplied robots.txt the RFC 9309 way for an agent deciding whether to fetch up to 25 paths as one product-token user agent: group selection (merged groups, "" fallback), longest-match with the Allow-wins-tie rule, "" and "$" patterns, percent-encoding normalization, fetch-status handling (4xx = no rules, 5xx = disallow all), the rule and line that decided each path. It also lints a suppl

Breaking-change diff between two JSON Schemas or OpenAPI documents, and whether cached calls still validate — Compares an old and a new JSON Schema 2020-12 (or two OpenAPI 3.0/3.1 documents) and lists each change with its path, keyword, old and new value, whether it narrows or widens what is accepted, and whether it breaks clients: newly required properties, removed or retyped properties, removed enum values, tightened limits, additionalProperties flips, removed operations, removed or newly required param

Repair a JSON value toward a schema using only the coercions the caller declares — Validates a JSON value against a caller-supplied JSON Schema 2020-12 and applies only the coercions you list (for example string-to-integer for "42", trim-strings, enum-case, apply-defaults, drop-additional) when each one strictly reduces the number of schema errors. Every change is returned with its JSON pointer, the rule, the old and new value; every error that remains is returned with its exact

One valid and one invalid example for each constraint in a JSON Schema — Walks a caller-supplied JSON Schema 2020-12 and, for every constraint keyword (type, enum, const, minimum, maximum, exclusiveMinimum, exclusiveMaximum, multipleOf, minLength, maxLength, pattern, format, minItems, maxItems, uniqueItems, required, minProperties, maxProperties, additionalProperties false, dependentRequired), returns the JSON pointer of the constraint, an example that satisfies it and

Score state for tennis, golf handicap differentials and cricket overs — Replays a tennis point sequence (ITF Rules of Tennis: games, deuce, tie-breaks, match tie-break, who serves), computes a golf Score Differential and Handicap Index (World Handicap System), or tallies a cricket innings ball by ball (overs.balls, wides and no-balls, maidens, wickets), for agents that must verify a reported score. Optionally checks a claimed score against the replay and names the fir

Batch Token Trap Screen — Runs the on-chain bytecode honeypot / rug screen on up to 10 Base tokens in one paid call, at 40% less than 10 single calls. Tokens the screen rejects as invalid input are reported per token; a chain or upstream outage fails the whole call so the payment is not settled.

Version range check: npm, PEP 440, Cargo, Maven — Parses a dependency version range in npm (node-semver), PEP 440 specifier, Cargo requirement or Maven range syntax and tests up to 100 candidate versions against it, for agents that write or review manifests. Returns per-version satisfies/violates with the failing comparator, the normalized range, the sorted versions and the highest match; pre-release, caret on 0.x, epoch and local-version rules f

Placeholder values hiding as data: caller-declared sentinels and likely candidates — Scans columns of values for sentinels. Values you declare in a dictionary (for example -1 means unknown in column age) are reported as asserted hits with their row paths. Other values that look like missing-data markers are reported separately as candidates, never asserted: the pandas default NA strings, spreadsheet error values, whitespace-only text, placeholder words, zero and epoch dates, fixed

Integer Sequence Identifier — Identifies an integer sequence from terms (a JSON array or pasted text), an OEIS A-number, a safe-parsed formula or recurrence, or a description, and leads with a verdict, confidence, A-number, formula, next terms and a next step. Searches the OEIS directly, then superseeker-style (differences, partial sums, gcd, ratios, binomial, Euler, leading terms, plus a search without the last two terms to f

Settlement date for a trade date: US equities T+1 on NYSE and Federal Reserve days, or T+n on a caller calendar — Computes the settlement date for a trade date under US equities T+1 (17 CFR 240.15c6-1, compliance date 2024-05-28) counting only days when the NYSE and the Federal Reserve Banks are open, or T+n on a named or caller calendar, and checks an asserted settlement date. The settlement-day rule (NYSE and Reserve Banks both open, so Columbus Day and Veterans Day are skipped) is our inference, not a publ

Is this lot still within shelf life on a given date? (unopened and after-opening limits, minimum remaining days) — For up to 100 lots, computes the expiry date from manufacture/fill date + shelf life (days, weeks, months, years; month arithmetic clamps to month end) and from opened date + after-opening life, takes the earlier, and checks it against a caller asOf date and an optional minimum remaining days at receipt; flags a labelled expiry that disagrees with the computed one. The last day of life is the expi

Shell quoting for bash, PowerShell and cmd.exe with a parse-back check — Turns up to 50 literal strings into a safely quoted argument for bash/sh (POSIX single-quote method), PowerShell (single-quoted literal) and cmd.exe, for agents that build a command line from untrusted or free-form text, and parses each result back with a model of that shell's rules to prove it yields the original string. It returns unsupported for strings cmd.exe cannot carry safely (line break

Site access: caller access rules against observed site facts — Checks whether an intended action on a website (read, crawl, submit, log in, ...) satisfies the access rules the caller supplies (robots rule read, terms clause, login present, rate-limit budget, captcha seen) against facts the caller observed, and says which rule blocked it or which fact is missing. For an agent deciding whether to proceed or ask. It is a reading of the caller's rules, not legal

Common free time across participants: intersect availability windows, report who blocks — Intersects up to 20 participants' free windows (up to 200 each, half-open RFC 3339 intervals with Z or offsets, so mixed time zones are fine) into the common slots at least minMinutes long, optionally clipped to a search window; when nothing fits, names the participants whose removal would create a slot. Built for an agent scheduling a meeting from already-fetched availability. It does not read ca

SMT Contract & Policy Logic Prover — Answers a yes/no logic question about contract clauses, IAM/OPA policies or business rules with a solver-backed result and a plain-language headline and next step. Takes clauses or pasted text (or a formal IR directly with zero LLM calls), formalizes them into typed Boolean, integer and rational constraints, then answers "is P possible", "does invariant I always hold" or "are any clauses contradic

Sun position, sunrise and sunset for a place and instant (NOAA equations) — Computes sun elevation, azimuth, declination, equation of time, sunrise, solar noon and sunset (UTC) for a latitude/longitude and up to 100 caller-supplied instants using the NOAA Global Monitoring Laboratory solar calculator equations (Meeus). Polar day and polar night return sunrise and sunset as null with a reason, not an error. Reads no clock: every instant comes from the caller. Does not mode

SPDX license expression parse and policy check — Parses up to 50 SPDX license expressions (AND, OR, WITH exceptions, + or-later, LicenseRef) and evaluates each against a caller policy (allow list, deny list, allowed exceptions, OSI/FSF flags, copyleft tier) for agents vetting dependency licenses. Returns the canonical expression, per-license flags from the pinned SPDX License List, the verdict, and which licenses satisfy an OR. Does not detect l

Normalize numbers, currency, phone numbers and units for text-to-speech — Rewrites digits in up to 20000 characters of text into spoken words for a TTS engine, for locales en-US and en-GB: cardinals, ordinals, years (with a preceding cue such as "in" or a month) and decades, decimals, percentages, currency with ISO 4217 minor units ("$12.50" -> "twelve dollars and fifty cents"), phone numbers read digit by digit, and SI/common units with locale spelling (meter/metre). O

Spend ledger with reservations: one budget for a tree of agents — A budget shared by a task tree, in integer minor units. reserve() atomically holds an amount if the remaining budget covers it (expired holds are freed automatically); commit() records what was actually spent, up to the hold; release() frees a hold. These are real reservations, not a snapshot: two workers cannot both reserve the last dollar. It records spending; it does not move money.

SQL guard: single statement, read-only, table allowlist — Checks one SQL string (up to 20000 characters; PostgreSQL, MySQL, SQLite or generic) before an agent executes it: exactly one statement, read-only (SELECT/VALUES/SHOW/EXPLAIN without side effects), and every table it reads or writes on a caller allowlist. Tokenizer-level: comments, string literals, quoted identifiers and dollar quoting cannot hide a second statement; reports statement kind, refere

State Machine Model Checker — Returns PASS, FAIL or INCONCLUSIVE for a finite state machine, protocol, or concurrent routine, with the shortest failing path as readable process.transition steps and a next step. Accepts a formal spec (no LLM), a plain-language workflow, or source code (translated to a spec by an LLM, with every transition quoted from the source and two independent translations compared). A pure TypeScript expli

Storage, demurrage or detention charge: free-time end date, chargeable days and tiered amount — Counts days between a received date and a released date for storage, demurrage, detention or parking charges: you state whether the receipt day and the release day count, how free days are counted (calendar or business days on a named or caller calendar), and a tiered per-day rate schedule by chargeable day number; returns the last free day, chargeable days, a per-tier breakdown and the total in m

Subtitle lint: SRT and WebVTT overlaps, duration, reading speed, line length — Parses an SRT or WebVTT file (up to 400000 characters, 4000 cues) and checks every cue for an agent about to deliver or burn in captions: end after start, overlaps and out-of-order cues, minimum gap, minimum and maximum duration, lines per cue, characters per line and characters per second (markup tags are not counted). Rules come from a named profile (netflix-en-adult, netflix-en-children, values

SVG layout check: clipping, overlaps and empty shapes from supplied geometry — Checks up to 200 SVG shapes you describe (rect, circle, ellipse, line, polygon, polyline, text, or a path's bounding box) against the viewBox and viewport the way an SVG renderer would: finds shapes clipped or entirely outside the drawn region, shapes that render nothing (zero or negative size, unstroked line), partial overlaps between shapes, and a zero, negative or malformed viewBox, for agents

Syllable counts per line, caller dictionary first, labelled English heuristic second — Counts syllables per word and per line for English text, using your dictionary entries first and otherwise a rule-based estimator (vowel groups, silent e, -ed/-es, consonant+le). Optionally checks each line against a target pattern such as 5-7-5. Where the rules cannot decide (vowel pairs like ia/io/eo, acronyms, numerals, non-Latin scripts) the word is a range or abstains, and a line whose count

Is this export in the npm package's published type declarations — Checks that up to 20 named exports (functions, classes, constants, types, default) exist in the .d.ts files published for an exact npm package version, read from jsDelivr, for coding agents that may be calling an API that does not exist. Follows export lists and relative export-star re-exports within the package (at most 7 files) and reports when the export surface cannot be fully enumerated. Neve

Detect near-duplicate passages (syndicated or copied text) so repeated sources are not counted as independent — Compares up to 30 passages an agent collected and reports which pairs share word-shingle content (Jaccard and containment), grouping them into clusters of probable copies so the number of independent sources can be stated honestly. Does not establish who copied whom, does not detect paraphrase or translation, and does not fetch anything; a high similarity is a hint that sources are not independent

Global Customs & Tariff Engineering Optimizer — Classifies a product under the HTS with GRI reasoning and verifies the code and general (MFN) rate against the live USITC schedule, flags when a specific sibling subheading may outrank the chosen residual one, separates lower-duty options from reclassification risk with dollar exposure, lists lawful physical tariff-engineering changes, and drafts an accurate commercial invoice description. Leads w

VAT / tax-id format and check digit (AT, BE, DE, DK, FI) — Checks the structure and check digit of up to 100 EU VAT numbers per call (Austria, Belgium, Germany, Denmark, Finland), for an agent about to put a supplier or customer VAT number on an invoice or into a record. Every other country answers unsupported rather than guessing a pattern. It never says whether the number is registered or active (that needs VIES or the national register), and a pass is

Are these timing constraints satisfiable? Earliest/latest time per event, or the conflicting constraints — Solves a simple temporal network: up to 40 named time points, up to 200 difference constraints 'b is between min and max minutes after a' and optional fixed instants, relative to a caller origin instant. When consistent it returns each point's earliest and latest feasible instant (where bounded); when not, it returns the cycle of constraints that contradict each other, with the total impossible sl

Interpret a test run: pass, no-tests, skipped, flaky, infra-failure from runner, exit code and counts — Reads a finished test run (runner, exit code, optional version, flags, caller-parsed counts, optional go test output, optional per-attempt exit codes) and says whether tests actually passed, ran nothing, were all skipped, passed only on retry, or never ran properly (timeout, signal, internal error). Encodes only exit-code semantics the runners document: pytest 0-5, node:test, mocha (failure count

Convert text offsets between UTF-16, code points, UTF-8 bytes and graphemes, and locate hidden Unicode — Reports the length of a string in UTF-16 code units, Unicode code points, UTF-8 bytes and extended grapheme clusters (Intl.Segmenter), converts up to 200 ranges given in any one of those units into all four, and flags ranges that fall inside a surrogate pair, a multi-byte UTF-8 character or a grapheme cluster. It also runs the unicode-check rules (bidi controls, zero-width and invisible characters

DST hazards for hourly and daily buckets in a zone: skipped hours, repeated hours, 23- and 25-hour days — For a time zone and a date range of up to 1100 days, lists every offset transition and what it does to local bucketing: the hourly labels that never occur (spring-forward), the hourly labels that occur twice and would merge two real hours when grouped by local time (fall-back), and the 23-, 24.5- or 25-hour days; optionally checks up to 500 caller hourly labels (YYYY-MM-DDTHH:00) for non-existent

SMPTE timecode conversion: drop-frame and non-drop, frames, exact seconds — Converts up to 200 timecodes, frame counts or times between SMPTE HH:MM:SS:FF (non-drop), HH:MM:SS;FF (drop-frame, 29.97 and 59.94), frame numbers and exact rational seconds, for an agent editing video, captions or logs at a stated frame rate. Rates are integers, a rational like 30000/1001, or the named 23.976, 29.97, 59.94, 119.88; any other decimal is rejected, never rounded. Integer arithmetic

Shared token bucket: many agents, one upstream rate limit — One rate limiter shared by every agent in your fleet. take(cost) atomically spends tokens if available and returns allowed plus remaining, or refuses with retryAfterMs until enough tokens refill. capacity and refillPerSecond are fixed when the bucket is first used; a call with different settings is refused so two fleets cannot silently disagree. Use a requestId so a retried take is not charged twi

Token Due Diligence Pack — Honeypot / rug screening from live Base bytecode plus a Solidity audit of the token's verified source. Source is fetched from Sourcify (following EIP-1967/zeppelinos proxies to the implementation) when not supplied; unverified tokens are refused before any charge. Runs token-trap-detector ($0.25) + contract-auditor ($1.00) in one paid call for 20% less than buying them separately. Charged only whe

On-Chain Bytecode & Unverified Token Trap Detector — Reads live Base bytecode, proxy slots (EIP-1967/beacon/zeppelinos/EIP-1822/EIP-1167), ERC-20 state, and privileged selectors to score honeypot and rug risk in code: SAFE | CAUTION | HONEYPOT_RUG | INSUFFICIENT_EVIDENCE. Adds honeypot.is buy/sell simulation and GoPlus tax and honeypot flags, disclosed per source, and degrades to a lower-confidence bytecode-only score when they fail.

Token and Vault Exit Ladder — One decision on holding a Base token and a vault position: token-trap-detector (live bytecode, proxy slots, honeypot simulation) and vault-exit-lawyer (redeem limits, queue, admin powers, eth_call simulation) run in parallel; contract-auditor then audits the token's verified source at the proxy implementation the trap step read, and is skipped on a honeypot or unverified source. Returns the worst

One-dimensional tolerance stack-up (worst-case and RSS) — Computes the minimum and maximum of a linear dimensional stack (the gap) from signed contributions with symmetric, unequal-bilateral, unilateral or min/max limits, both by exact worst-case arithmetic and by root-sum-square, names the dominant contributor and tests the gap against a required minimum and/or maximum. For mechanical designers and agents checking clearances. It does not do 2D or 3D sta

Transcribe an audio or video file URL (Whisper, timestamps, SRT, VTT) — Transcribes a public audio or video FILE URL (podcast enclosure, meeting recording, voice memo; mp3, wav, mp4/m4a, ogg, webm, flac; up to 25 MB and 30 minutes) with Whisper large-v3-turbo on Cloudflare Workers AI. Returns timestamped segments plus SRT and VTT subtitles and the plain text, bound to the media's SHA-256. Does NOT accept YouTube, Spotify or any streaming page (the URL must be the medi

Check tool-call / tool-result pairing and ordering in an LLM transcript (Anthropic, OpenAI Chat, OpenAI Responses) — Validates a message list before it is sent to Anthropic Messages, OpenAI Chat Completions or OpenAI Responses: every tool call paired with its result, results in the right place and role, ids present and unique. Built for an agent that edits, truncates or replays its own history and would otherwise get a 400. Only rules stated in the providers' documentation fail the check; anything else suspiciou

Check that numbers, units, part numbers, tolerance signs, dates, placeholders and URLs survive translation — For aligned source and target segments, extracts the tokens a translation must not change (numbers with attached units or %, ISO dates, part numbers and versions, tolerance and comparison operators such as ± ≤ ≥, placeholders like {0} or %s, URLs and emails) and reports any that are missing, added or altered, with offsets in both texts in UTF-16, code points, UTF-8 bytes and graphemes. Numbers are

Unicode smuggling check for strings (prompts, tool args, URLs) — Checks up to 100 strings in one call against the 19 deterministic rules of unicode-safety-check: bidi controls (Trojan Source), invisible and zero-width characters, tag-character and variation-selector payloads, homoglyph and mixed-script hosts or names. Built for an agent's hot path: run it on a prompt, a tool argument, a URL or a model output before acting on it. Returns BLOCK / REVIEW / NO_FIND

Adversarial Unicode & Trojan Source gatekeeper — Scans a public GitHub repo, a pull request, a unified diff or a list of files for adversarial Unicode with the 19 rules (USC001-USC019) of dcondrey/unicode-safety-check: bidi controls (Trojan Source), invisible and default-ignorable characters, variation-selector and tag-character payloads (Glassworm), private-use code points, homoglyph and mixed-script identifiers, confusable collisions, misplace

Which oneOf / anyOf branch a JSON value matches, or why several or none do — Evaluates a JSON value against every branch of a oneOf or anyOf in a caller-supplied JSON Schema 2020-12 (or OpenAPI schema converted to it) and reports which branches match, the exact errors of each branch that fails with the JSON pointer of each offending value, the closest branch, and which branch an OpenAPI discriminator points to. For agents debugging why a tool or API payload is rejected w

Exact unit conversion with dimensional analysis (NIST SP 811) — Converts up to 100 values between units for an agent that needs the exact factor and a refusal when the units do not match: SI and NIST SP 811 length, mass, time, volume, pressure, energy, power, temperature, with compound expressions like kg*m/s^2. Rejects incompatible dimensions, converts absolute temperatures affinely through kelvin and temperature differences linearly, and converts mass to vol

What do omitted, null, [] and "" mean in this PATCH body under merge-patch, JSON Patch or a declared dialect? — For an agent about to send a PATCH or update body: states, per field, what omitted / null / [] / "" mean under the declared dialect (RFC 7396 merge patch, RFC 6902 JSON Patch, or caller-declared semantics) and checks the body against stated intents such as 'clear this field', 'leave this alone' or 'set to empty array'. With a target document it applies the patch and compares the result; without on

URL canonicalizer and lookalike-host check — Canonicalizes up to 50 absolute http(s) URLs for an agent about to fetch, dedupe or click them: WHATWG-parsed lowercase host, default port and fragment and userinfo removed, dot segments and unreserved percent-escapes normalized, tracking parameters (utm_*, gclid, fbclid and the Mozilla strip list plus yours) removed, and flags userinfo@host tricks, backslash and multiple-@ authorities, obfuscated

Vault Exit Lawyer — Before moving treasury funds out of an ERC-4626 or ERC-7540 vault: reads the chain at one pinned block and returns how much can exit now, the haircut vs convertToAssets, partial-exit maximum, liquidity coverage, queue/cooldown timing, admin-power flags (upgrade, pause, multisig threshold, timelock), and a verdict EXIT_NOW_FULL / EXIT_PARTIAL / QUEUED / BLOCKED / UNKNOWN. With a holder it simulates

VIN decode: manufacturer (WMI), model year, check digit — Decodes up to 50 VINs per call for an agent handling a vehicle listing, claim or order: validates the 17 characters and letters I/O/Q, the position-9 check digit (49 CFR 565.15), the world manufacturer identifier from a bundled NHTSA vPIC snapshot, and the model year from position 10. Position 10 repeats every 30 years: pass vehicleClass light (position 7 resolves the century) or asOf, otherwise b

WARM verifiable agent memory — Persistent memory for AI agents as a pay-per-call API, no database to run. op=remember stores subject-predicate-object facts (or extracts them from free text; only facts whose subject and object appear verbatim are kept); a newer value for the same subject and predicate supersedes the old one, and the history stays queryable. op=recall searches by BM25 keyword ranking or exact triple pattern, now

Verify webhook signatures from 8 providers — Constant-time verification of inbound webhook signatures from Stripe, GitHub, Shopify, Slack, Twilio, Linear, Clerk (Svix) and SendGrid HMAC, each using the exact signing string that provider documents. Every scheme uses Web Crypto HMAC with a timing-safe compare and, where the provider signs a timestamp, a configurable clock-skew tolerance (default 300s) to block replay. Returns the verdict and t

Web search: licensed Exa/Brave results with snippets — Runs a web search through a licensed search API (Exa or Brave) and returns up to 10 ranked results with title, URL, snippet, highlights and published date, for agents that need current links. Does not scrape search engines, does not fetch or summarize the result pages (use page-contents for that), and does not use a model.

Week number and week-year for dates under ISO 8601, US or custom rules (53-week years) — Returns the week number, week-year, number of weeks in that year and the week's first and last day for up to 200 dates under ISO 8601 (Monday start, week 1 holds 4 January), the US convention (Sunday start, week 1 holds 1 January) or a custom weekStart/minimalDays rule, and checks an asserted week or week-year (2027-01-01 is ISO 2026-W53). Built for an agent reading or writing week-based dates in

Is there a compatible PyPI wheel for this Python and platform — Lists the wheels PyPI offers for a package release and picks those compatible with a given CPython version, OS, CPU architecture and libc (glibc or musl version, or macOS version), best match first, for agents choosing base images or debugging 'building from source' installs. Applies PEP 425/600/656 tag ordering, abi3, legacy manylinux aliases, Requires-Python and yanked files. CPython only (no Py

Is this state transition allowed by the caller's workflow definition, and what is allowed instead? — Given a caller-supplied state machine (transitions with optional event names and required facts), the current state and a requested event or target state, says whether the transition is allowed, which required facts are false or unknown, and which transitions are available from the current state. For an agent about to move an order, ticket, deployment or approval to its next state. Terminal states

X.509 certificate inspection from PEM — Takes a PEM-encoded X.509 certificate and returns its fields (subject, issuer, serial, validity window, SANs, key usage, EKU, basic constraints, signature algorithm, public key algorithm and bit size) plus a fail-on-issue audit: expired or not-yet-valid, hostname/SAN mismatch per RFC 6125 §6.4.3 wildcards, weak signature (SHA-1), short RSA (<2048), serverAuth without SAN (CABF BR 7.1.4.2.1). Decod