agent-doc-skillsCAUTION
Navigable project documentation for Claude Code, Cursor, and Codex — a knowledge base your coding agent can pick up cold, instead of rediscovering the repo every session.
Overview
From the repository's own README, as read at the audited commit. Badges and raw HTML are left out.
Languages: English | 简体中文
Three Agent Skills keep project knowledge discoverable without loading every document into every conversation. Skills retain detailed procedures in linked references and use small task entrypoints.
Install
On macOS, Linux or Windows, with Node.js installed:
npx skills add x0c/agent-doc-skills -g
The installer offers doc-init, doc-compact and doc-update for compatible Agent Skills hosts. Python 3 is required for the bundled helpers. Start a new conversation after installation.
First use
Use doc-init to establish documentation for this project from its existing authorities.
For an existing project, choose the skill matching the work:
For a read-only review, say Use doc-compact to audit this project's documentation without changing files. For routine maintenance, say Use doc-update to record this task's reusable findings in their owning documents.
Operating boundaries
Record adopted durable decisions before implementing them. Use accurate content summaries so an ordinary task can discover its relevant standards before a design decision. Preserve scope, exceptions, evidence, numbers and machine-parsed markers during compression.
The current documentation preset is v21. Its injector discovers real global instruction files and preserves neighboring user and managed content. It does not inject the global preset into product instruction files. Root AGENTS.md edits use the s
7e65c8a1dd13OBSERVED · 2026-10-12Install
Commands as the repository documents them. They are shown, not run.
npx skills add x0c/agent-doc-skills -g
npx skills add x0c/agent-doc-skills -g
Host compatibility
What the documentation claims. We have not run a compatibility test.
| Host | Status | Notes |
|---|---|---|
| claude-code | mentioned | |
| codex | mentioned |
What it tells the agent
The instruction file, verbatim from the audited commit — this is the text the model reads, and the surface the audit's instruction layer examines. Quoted here so you can judge it without cloning anything.
--- name: doc-compact description: Audit and compress bloated project documentation while preserving behavior and repairing navigation. Use for redundant docs, broken indexes, oversized AGENTS.md, malformed CLAUDE.md, or missing global documentation governance. --- # Documentation compaction Audit and compress documentation while preserving adopted behavior, evidence and discoverability. Follow the installed lifecycle and environment policies; use `agents-md-edit` for AGENTS.md edits. Delegate only under current user authorization. ## Choose the work - **Read-only audit:** run `python3 scripts/audit.py PROJECT_ROOT` and inspect actual coverage, misplaced/repeated content and relevant authority. Use the audit sections of [maintenance workflow](references/maintenance-workflow.md#step-2--read-only-audit). Do not install presets, stamp or repair files. - **Bounded cleanup:** inspect the affected domain and overlapping references, then use the relevant structure/compression/verification sections of that workflow. Do not expand a small request into a repository census. - **Whole-project cleanup:** read [maintenance workflow](references/maintenance-workflow.md). Discover effective global instructions, audit each actual documentation host, decide structure by readability, compress content and verify the full requested scope. ## Preservation and verification Change generated or inherited rules through their actual source; preserve managed markers and valid coverage. Use concise content summaries with the shared relevance instruction, without arbitrary reading quotas. Compression does not drop scope, exceptions, timing, evidence or machine-parsed literals. Read [compression guide](references/compression-guide.md) before substantive compression. Distinguish a smaller file from preserved behavior. Check navigation, source/path liveness, ordinary and exceptional scenarios, and required compact-stamp coverage. Report per-domain results and unresolved factual contradictions; a successful script or stamp is limited coverage evidence. ## Migration completion and stop rule An authorized knowledge migration is complete when the shared authority preserves the reusable knowledge's applicability and exceptions, local product contracts and evidence remain intact, generic duplication is replaced where safe, and content rereading plus navigation checks reveal no unresolved material defect. For multi-file work, reuse one concise source → authority → local retention/outcome table when useful; a small edit needs no extra ledger. After accepted changes pass the scoped checks, close out rather than starting another harvest or wording pass. Reopen for a concrete defect, contradiction, missed authorized item, or new user direction. Additional reviews need a specific unresolved question; optional polish alone does not require another cycle.
Trust audit
CAUTIONgrade C · trust 79/100 Install with care. The audit found things worth knowing before you trust its output.
| Layer | What it checks | Result |
|---|---|---|
| L0 | Provenance & inventory | PASS |
| L1 | Static analysis of the code | PASS |
| L2 | Instruction surface (what it tells the agent) | FAIL |
| L3 | Class-specific surface | PASS |
| L4 | Behavioural (sandbox) | SKIPPED |
What the source does
- Filesystem
- none-observed
- Network
- none-observed
- Shell
- none-observed
- Dependencies
- pinned
- Secrets in source
- none-found
Findings (4)
- On connection failure, insufficient permissions, or suspected high-risk production access, report missing items; do not bypass safety limits.
- API validation (HTTP scenarios): `curl -s -X POST http://127.0.0.1:<port>/<path> -H 'Content-Type: application/json' -d '<minimal body>'`
- Cache/state refresh (caching scenarios): `curl -s -X POST http://127.0.0.1:<port>/<refresh_path>`
When the user explicitly says “don’t ask,” “test mode,” “dry run,” “report only,” “don’t write files,” or “limit scanned files / docs / budget,” enter convergent mode and strictly respect those bounds
Gates applied: instruction_override, no_behavioural_pass.
7e65c8a1dd13full audit observations/trust-audit/skill/x0c__agent-doc-skills.json · Report an issue / request a re-scanAudit history
Every audit this skill has had.
| Date | Source | Verdict | Grade | Score | Change |
|---|---|---|---|---|---|
| 2026-10-12 | 7e65c8a1dd13 | CAUTION | C | 79 | first audit |
Questions
What does the agent-doc-skills skill do?
Navigable project documentation for Claude Code, Cursor, and Codex — a knowledge base your coding agent can pick up cold, instead of rediscovering the repo every session.
Is agent-doc-skills safe to install?
With care. The audit graded it C (79/100) and found 4 things worth knowing before you trust this skill, listed below with the exact line each was found on.
What can agent-doc-skills access on my machine?
The audit observed no filesystem, network or shell use at all in its source.
Which assistants does agent-doc-skills work with?
Its documentation mentions claude-code and codex. That is what the text claims, not a compatibility test we ran.
How current is this page?
The grade is for one exact copy of the source (7e65c8a1dd13), read on 2026-10-12. The repository is watched, and a new audit runs when it changes — this is the first audit.