mcpField
C
compound-docs

compound-docs

github.com/iliaalDocuments45140 installsUpdated 4d ago
Repository

About compound-docs

compound-docs is a skill for capturing resolved engineering problems as structured Markdown documentation. It guides the process from confirming a reusable lesson and checking existing entries to validating metadata and linking related issues.

Key features

  1. 01Capture confirmed, reusable fixes
  2. 02Collect root cause and prevention details
  3. 03Check for existing solution documents
  4. 04Validate YAML frontmatter
  5. 05Cross-reference related issues

Use cases

  1. 01Record a fix that could otherwise be repeated
  2. 02Update or link documentation for similar issues
  3. 03Review recurring problems for critical patterns

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/iliaal/ai-skills --skill compound-docs

After installing, type/compound-docsin Claude Code to confirm it loaded.

Bundled files

References and scripts installed along with SKILL.md.

SPEC.md
agents/
openai.yaml
assets/
critical-pattern-template.md
resolution-template.md
references/
documentation-process.md
example-scenario.md
quality-guidelines.md
yaml-schema.md
scripts/
validate-frontmatter.py
validate-frontmatter.sh

Details

From the project README

compound-docs

Process

Single-file architecture -- one markdown file per problem in its symptom category directory (e.g., docs/solutions/performance-issues/n-plus-one-briefs.md), with YAML frontmatter for metadata.

Follow the 7-step documentation capture process. For full details, see documentation-process.md.

  1. Detect confirmation -- Auto-invoke after "that worked", "it's fixed", etc. Skip trivial fixes.
  2. Gather context -- Extract module, symptom, investigation attempts, root cause, solution, prevention. BLOCK if critical context missing.
  3. Check existing docs -- Search docs/solutions/ for similar issues. If found, offer: new doc with cross-reference, update existing, or other.
  4. Generate filename -- Format: [sanitized-symptom]-[module]-[YYYYMMDD].md
  5. Validate YAML -- Save the draft to a temporary Markdown file, then run validate-frontmatter.sh against it. The validator requires Python 3 and PyYAML; report a missing dependency rather than claiming validation. If invalid, fix the frontmatter and re-run until it passes.
  6. Create documentation -- Write file to docs/solutions/[category]/[filename].md using resolution-template.md.
  7. Cross-reference -- Link related issues. Detect critical patterns (3+ similar issues).

Capture gate (step 1): capture only if, without this doc, a future engineer reading the final code, tests, comments, CLAUDE.md/AGENTS.md, and existing docs or skills would still repeat the mistake or redo the investigation. Skip when any of those already carries the lesson.


Decision Menu

After successful documentation, present the options below via AskUserQuestion (Claude Code; load with ToolSearch select:AskUserQuestion if not loaded) or request_user_input (Codex), falling back to the numbered menu in chat, and WAIT for the user's response:

Solution documented

File created:
- docs/solutions/[category]/[filename].md

What's next?
1. Continue workflow (recommended)
2. Add to Required Reading - Promote to critical patterns
3. Link related issues - Connect to similar problems
4. Add to existing skill - Add to a learning skill
5. Create new skill - Extract into new learning skill
6. View documentation - See what was captured
7. Other

For detailed response handling, see documentation-process.md.


Success Criteria

  • YAML frontmatter validated (all required fields, correct formats)
  • File created in docs/solutions/[category]/[filename].md
  • Enum values match schema exactly
  • Code examples included in solution section
  • Cross-references added if related issues found
  • User presented with decision menu and action confirmed

References

Integration

  • Stale-learning review (/ia-compound-refresh in Claude Code) -- reviews docs/solutions/ for entries that have aged out

Comments

Markdown supported