Detect which IDE skill/rule layouts exist in the current working directory. Returns the list of matching adapters (e.g. cursor, claude, codex, kiro). Local-only, no network.

List all skill / rule files found in the detected IDE directories under the current working directory.

Read a local skill file (raw contents). Path must be inside the current working directory.

Write a local skill file. Creates parent directories. Path must be inside the current working directory.

Lint a skill file: front-matter, token count, broken links, TODO markers. Local-only.

Validate a skill file against the agentskills.io standard.

Convert a skill file from one IDE format to another (e.g. cursor → claude). Round-trips through a canonical {frontmatter, body} representation.

Diff a local skill file against its cloud counterpart by slug. Local-side diff is computed here; the cloud half requires MODELBOUND_API_KEY.

Create a new SKILL.md with default scope constraints and draft review state.

Flight Check: offline safety scan for a skill or prompt (prompt injection, exfiltration, secrets, destructive commands, wildcard tools, loop hazards, clarity, token budget). Pass a file path or raw content. Local-only, zero AI cost. Blocking findings should be fixed before shipping.

Check a local skill with deterministic heuristics (scanner h5). Returns pass / warn / fail with findings — no numeric score.

Show review_state, review_meta, check status, and confidence trend for a skill.

Move a skill to pending_review.

Approve a skill; stores the approved body hash and checks status (pass, warn or fail) at approval time.

Reject a skill with optional reviewer notes.

CI gate: exit semantics via blocked=true when skill is not approved.

Pull a skill from the ModelBound cloud library. Response includes ai_type, source_platform, source_path, and repo when known. Requires MODELBOUND_API_KEY.

Create or update a skill in the cloud library. Requires MODELBOUND_API_KEY.

List skills in the cloud library. Supports ai_type (skill|hook|steering|system-prompt|rule|agent|memory|spec|instructions|prompt) and source_platform (claude-code|cursor|kiro|amazon-q|copilot|windsurf|codex|agents|modelbound) filters. Each row includes ai_type, source_platform, source_path, and repo for hierarchy-aware orchestration. Requires MODELBOUND_API_KEY.

Return the team's full AI resource hierarchy grouped by platform → top-level directory (.claude/skills, .cursor/rules, .kiro/steering, …) → files. Use this BEFORE listSkills when an orchestrator needs to map available context before loading. Requires MODELBOUND_API_KEY.

Full-text search across all cloud content. Requires MODELBOUND_API_KEY.

Install a public marketplace skill into your library. Requires MODELBOUND_API_KEY.

Get token health scores and staleness for your context library. Requires MODELBOUND_API_KEY.

Run AI token optimization on a cloud skill, file, agent, or pack. Returns tokens_saved, savings_pct, version snapshot id, and a per-file breakdown. Requires MODELBOUND_API_KEY. Server creates a backup version automatically; for local-file edits the caller MUST snapshot via the BackupEngine first.

Same as optimization.run but never writes. Returns projected savings. Safe to call from auto-suggest hooks. Requires MODELBOUND_API_KEY.

Fetch a cached optimized variant of a skill or file without re-running optimization. Useful when an agent wants the lean version for inference but the raw version for editing.

Set workspace context before skill list/get/test/pipeline ops. Scopes skills to the current repo by default. Requires MODELBOUND_API_KEY.

Sync a local skill file from the IDE to cloud and return a repo-linked skill UUID. Requires MODELBOUND_API_KEY.

Run the Skill Development Pipeline (Test → Optimize → Production) on a skill. Streams stage results into skill_pipeline_runs. Returns { run_id, status, version_after }. Use pipeline.status to watch progress. Requires MODELBOUND_API_KEY.

Get pipeline run status. Pass run_id for a specific run, or skill_id + limit for recent runs. Poll every 1-2s while status === 'running'.

Read or update a skill's pipeline_config (max_latency_ms, enforce_latency/enforce_tests flags, default targets). Pass config to update; omit to read.

Run a single skill test case against the live model. Pass test_case_id for a saved case or prompt + skill_md for an ad-hoc run. Returns adherence_score (1-10), verdict, latency_ms.

List saved test cases for a skill.

Create a saved automated check test case for a skill. Requires skill_id, name, and prompt. Optional notes for grading.

List recent test runs for a skill (latest first).

List historical versions of a skill or file (from file_versions). Each entry includes version label, sha, created_at, and the actor that produced it.

Return a unified diff between two versions of a skill (or between the live version and an arbitrary string). Use skill.versions first to find version labels.

List safety and quality findings for a skill. Returns findings with stable keys, and ignored_keys. Requires MODELBOUND_API_KEY.

Ignore a finding by stable key (or class/severity/message). Requires MODELBOUND_API_KEY.

Un-ignore a previously ignored finding. Requires MODELBOUND_API_KEY.

Run benchmark latency suite for a skill. Requires MODELBOUND_API_KEY.

Compare two skill versions (e.g. latest vs current). Requires MODELBOUND_API_KEY.

Suggest skill improvements from findings, tests and real-world outcomes. Requires MODELBOUND_API_KEY.

Create a new eval test case (name, input_prompt, optional expected_output, rubric). Requires agent scope.

List all eval cases for the team. Requires agent scope.

Submit actual output for scoring against an eval case. Pass score/pass for manual judging or omit for AI judge.

List past eval results, optionally filtered by eval_case_id.

Report how a skill performed after you used it. Call this once the task that used the skill is finished — it is how the skill gets improved. Keep excerpts short. Requires MODELBOUND_API_KEY.

Read back how the team's skills have performed in real use (worked / partial / failed over a rolling window). Use it to pick the most reliable skill for a task, or to find the ones that need attention. Requires MODELBOUND_API_KEY.

Check whether a skill is cleared to run unattended. Returns a pass/warn/fail status across the four harness pillars (context, permissions, guardrails, verification) from the skill's latest pipeline run, with the reasons for anything blocking. Call this before using a skill for a task you will not review.

Report a finished agent run to ModelBound so each step is tied to the skill that shaped it. Send step summaries only (name, kind, duration, tokens, status) — never prompts or outputs. Failed runs become outcome reports for the skill.