Read and analyze PDF, DOCX, Excel, or PowerPoint (.pptx) files. Modes: 'summary' (overview with preview), 'indepth' (full text, structure, metadata), 'focused' (query-based search). For a .pptx it returns a per-slide transcript (titles, bullets, speaker notes) and the slide count. Source: either local filePath OR a remote https url whose response is {data:base64, filename, mimeType, size} JSON guarded by authHeader (used for one-shot capabilities like CogniRunner attachment fetches). Always read before editing; use 'indepth' before edit-doc.

PLAN the best output format BEFORE creating. Call this FIRST whenever the user didn't explicitly name a format. It weighs explicit format words, what the user wants to DO with the file, topic, and content shape — then returns a ready-to-use creation plan. Nuance it captures: 'README / API / spec / for the repo' → markdown; 'budget / tracker / dataset / table' → excel (CSV if they say csv); 'editable / draft / template / in Word' → docx; 'print / send to the client / official / invoice / resume / final / sign' → PDF. DOCX = editable Word; PDF = final, fixed-layout, print/sign/send. Returns { format (markdown|docx|excel|pdf), suggestedTool, stylePreset, category, docType, confidence, reason, alternativeFormat, outputFormat? ('csv'), unsupported? ('pptx'), note? }. Pass these straight into the create-* tool. There is no native slides/PowerPoint tool yet — it recommends the closest fit (usually PDF) and flags unsupported:'pptx'.

Create a styled, EDITABLE Word DOCX. USE for stakeholder/business/legal/research deliverables the user will keep editing in Word, or when they say 'Word / .docx / editable / draft / template'. NOT for: a final/print/send-as-PDF deliverable (→ create-pdf), code/API/README docs (→ create-markdown), or tabular/numeric data (→ create-excel). The most full-featured tool: Document DNA defaults, 8 style presets, headers/footers with page numbers, margins, blueprint validation, and real tables. ALWAYS format the body with markdown — never a wall of plain text. Simplest: put the whole body in the content string. Supported markdown: '# H1' '## H2' '### H3' headings; 'bold'; 'italic'; 'code'; '- ' or '1. ' lists; '> ' blockquotes; '---' horizontal rule; fenced code blocks; '| a | b |' GitHub tables (with a '|---|---|' separator row); 'text' links. EXAMPLE content: "## Overview\nThis report covers Q2 results.\n\n### Highlights\n- Revenue up 18%\n- Two new markets\n\n| Metric | Value |\n|---|---|\n| MRR | $42k |\n| Churn | 1.2% |\n\n> Next review: July." Title MUST be specific. Duplicate → { duplicate: true, existingPath } (switch to edit-doc). Response includes formattingQuality and formatSuggestion — if formatSuggestion is set, the content fits another format better, so heed it. Use dryRun: true for preview.

Create a Markdown (.md) file for technical/code content that lives in a repo. USE for READMEs, API docs, specs, runbooks, changelogs, integration guides, code-heavy content — anything for GitHub/developers. NOT for stakeholder-facing or printable deliverables (→ create-doc / create-pdf) or tabular data (→ create-excel). MARKDOWN SUPERPOWERS: set toc: true to auto-generate an anchor-linked Table of Contents from the H2/H3 headings; pass frontmatter: {...} to emit YAML frontmatter (title, date, tags[]) for static-site generators (Hugo/Jekyll/Astro). Simplest usage: put the whole body in the content string. Supported markdown: '# H1' '## H2' '### H3' headings; 'bold'; 'italic'; 'code'; '- ' or '1. ' lists; '> ' blockquotes; '---' horizontal rule; fenced code blocks; '| a | b |' GitHub tables (with a '|---|---|' separator row); 'text' links. The title becomes the H1, so start content at '## '. Title MUST be specific. Response includes formattingQuality. Use dryRun: true for preview.

Create a styled Excel XLSX (or CSV). USE for ANY tabular/numeric data — budgets, trackers, datasets, KPIs, price lists, schedules; even a table inside a 'report' belongs here, not in a doc. EXCEL SUPERPOWERS: a cell whose string starts with '=' becomes a LIVE formula (e.g. "=SUM(B2:B9)", "=B2C2") that Excel computes on open; columns whose header looks like money ($/price/cost/revenue) or percent (%) auto-format; the header row gets autofilter dropdowns; columns auto-fit. The first row of each sheet is the header (auto-styled bold/filled), body rows get zebra striping — you only supply values. Set outputFormat: "csv" for a plain CSV (first sheet only; no styling/formulas). EXAMPLE: sheets: [{ name: "Q2 Revenue", data: [["Month","Units","Price","Revenue"],["Apr",120,9.99,"=B2C2"],["Total","=SUM(B2:B2)","","=SUM(D2:D2)"]] }]. Sheet names MUST be specific (rejects 'Sheet1', 'Data', etc.). Use dryRun: true for preview.

Create a FINAL, fixed-layout PDF to read / print / send / sign. USE when the user says PDF / print / 'send to the client' / official / invoice / flyer / resume / cover letter / 'read-only' / 'final version'. NOT for content they'll keep editing (→ create-doc) or code/repo docs (→ create-markdown). Rendered from markdown with the same 8 presets as create-doc, via headless Chromium. PDF SUPERPOWER: set toc: true for a clickable Table of Contents (with heading anchors) at the top. Supports headers/footers with {current}/{total} page numbers and margins. ALWAYS format the body with markdown. Easiest: put the whole body in the content string. Supported markdown: '# H1' '## H2' '### H3' headings; 'bold'; 'italic'; 'code'; '- ' or '1. ' lists; '> ' blockquotes; '---' horizontal rule; fenced code blocks; '| a | b |' GitHub tables (with a '|---|---|' separator row); 'text' links. EXAMPLE content: "## Overview\nThis report covers Q2 results.\n\n### Highlights\n- Revenue up 18%\n- Two new markets\n\n| Metric | Value |\n|---|---|\n| MRR | $42k |\n| Churn | 1.2% |\n\n> Next review: July." Title MUST be specific. Response includes formattingQuality and formatSuggestion. Use dryRun: true for preview. (To READ a PDF, use read-doc.)

Create an EDITABLE PowerPoint presentation (.pptx) you can open in PowerPoint / Keynote / Google Slides. USE when the user asks for slides / a deck / a presentation / a pitch deck / 'powerpoint' / 'keynote'. NOT for a flowing document (→ create-doc / create-pdf), code/repo docs (→ create-markdown), or pure tabular data (→ create-excel). SLIDE STRUCTURE: the title becomes a centered title slide; EACH '## ' heading starts a NEW slide whose body is the markdown beneath it. Inside a slide, '### ' is a sub-heading, '- '/'1. ' are bullets, '| a | b |' GitHub tables render as native slide tables, a chart fenced block (first line 'type: bar|column|line|pie|doughnut|area', optional 'title:', then a markdown table whose first column is the category and each other column is a data series) becomes a NATIVE editable chart, and fenced code blocks render as monospace. Keep each slide focused — a few bullets, not a wall of text. Styled from the same 8 presets as create-doc (colors/fonts map onto the slides). Supported markdown: '# H1' '## H2' '### H3' headings; 'bold'; 'italic'; 'code'; '- ' or '1. ' lists; '> ' blockquotes; '---' horizontal rule; fenced code blocks; '| a | b |' GitHub tables (with a '|---|---|' separator row); 'text' links. EXAMPLE content: "## Problem\n- Manual steps are slow\n- Errors slip through\n\n## Our Solution\n- One-click automation\n- Built-in checks\n\n## Results\n| Metric | Before | After |\n|---|---|---|\n| Time | 2h | 5m |". Title MUST be specific. Use dryRun: true to preview the slide breakdown. (To make a fixed, non-editable deck use create-pdf instead.)

Edit an existing PowerPoint (.pptx). Actions: 'preview' (show the current slide outline), 'append-slides' (add new slides from markdown — one '## ' heading per slide), 'replace-slide' (replace one content slide by 1-based index with new markdown). IMPORTANT: edit-pptx REBUILDS the deck from the existing slides' extracted TEXT + speaker notes, normalized to a style preset — charts, images, and exact original formatting on pre-existing slides are NOT preserved (best for the text/bullet decks create-pptx makes). To author a brand-new deck use create-pptx; to read a deck use read-doc.

Edit an existing DOCX. Actions: 'append' (XML-patches new content, PRESERVES original formatting/headers/footers/images), 'replace' (overwrites body but keeps section properties), 'style' (apply a stylePreset to existing paragraphs without changing text), 'preview' (show what would change). Always read-doc with mode 'indepth' first so you understand the existing structure.

Edit an existing XLSX workbook. Actions: 'append-rows' (add rows to a sheet, preserves existing styles), 'append-sheet' (add a new sheet — fails if name exists), 'replace-sheet' (overwrite a sheet's data), 'preview' (show what would change). Use read-doc first if you don't know the sheet structure.

Search the document registry. Filters compose with AND-logic: category match AND tag match AND title-substring match. Returns an array of registry entries with { id, title, filePath, category, tags, description, createdAt, updatedAt }. Use this before create-doc to check if a similar document exists.

List available document templates AND blueprints. Templates are static, named structures (e.g. 'claude-like', 'technical-docs', 'business-report') you can reference via tags on create-doc. Blueprints are auto-learned structures stored in .document-blueprints.json. Use the returned name as blueprint: in create-doc, or as a tag for create-doc tag-based style detection.

Manage the project's Document DNA — header/footer/style defaults that auto-apply to every create-doc call. Actions: 'init' (one-time setup with companyName/header/footer/stylePreset), 'get' (current config + project profile), 'evolve' (analyze usage trends; with apply:true, MUTATES dna config and may auto-create blueprints — irreversible without manual cleanup), 'save-memory'/'delete-memory' (project-wide preferences keyed by string).

Manage structural blueprints — section/heading templates extracted from real documents. Actions: 'learn' (extract from a DOCX or PDF you already have), 'list' (show stored blueprints), 'delete' (remove by name). Blueprints are also auto-learned during 'dna evolve' when recurring structures are detected. Use a blueprint by passing { blueprint: '' } to create-doc — the tool will validate that your paragraphs match the structure.

Monitor documents for structural drift over time. Actions: 'watch' (compute and store a fingerprint = SHA-256 + heading tree + word counts; capped at 500 paragraphs), 'check' (compare current state against the stored baseline; reports word-count delta, added/removed headings, category shifts, and a similarity score). Omit filePath on 'check' to compare all watched documents.

Trace the provenance chain of a document. Returns a tree { sources: [{filePath, ...}], derivatives: [{filePath, ...}] } showing which read documents informed this document (sources, traced upstream) and which created documents derived from it (derivatives, traced downstream). Lineage is recorded automatically when read-doc and create-doc are called within the same session.

Fact-check a document (or explicit claims) against the LIVE WEB. CROSS-MCP tool: doc-processor extracts the claims, then CALLS the web-search MCP (get-web-search-summaries) to gather sources per claim, and optionally writes a cited PDF report. Provide claims (an array of statements) OR a filePath/content to auto-extract factual claims from. Because this reaches the web-search MCP you MUST pass webSearchBearer (your web-search demo key) and serperKey (your Serper key); webSearchUrl defaults to the hosted web-search endpoint. Returns, per claim, the retrieved evidence + source URLs + a ROUGH keyword-overlap support score — NOT a verdict; read the evidence and decide support/refute yourself. Set generateReport: true for a downloadable cited PDF.