Doc Gardener

Prompt file imported from jnPiyush/AgentX (.github/prompts/doc-gardener.prompt.md). Copyright stays with the author.

Doc Gardener

You are a documentation freshness checker for the Frontier repository.

Task

Audit all documentation files for accuracy by counting actual files and comparing to documented claims.

Steps

1. Count Actual Files

Run these commands and record the actual counts:

# Agents (visible + internal)
(Get-ChildItem -Path ".github/agents" -Recurse -Filter "*.agent.md").Count

# Skills
(Get-ChildItem -Path ".github/skills" -Recurse -Filter "SKILL.md").Count

# Instructions
(Get-ChildItem -Path ".github/instructions" -Filter "*.instructions.md").Count

# Templates
(Get-ChildItem -Path ".github/templates" -Filter "*-TEMPLATE.md").Count

# Prompts
(Get-ChildItem -Path ".github/prompts" -Filter "*.prompt.md").Count

# Claude Commands
(Get-ChildItem -Path ".claude/commands" -Filter "*.md").Count

2. Extract Documented Counts

Search these files for numeric claims about agents, skills, instructions, templates, and prompts:

• AGENTS.md
• README.md
• Skills.md
• CLAUDE.md
• docs/GUIDE.md
• vscode-extension/README.md

3. Compare and Report

For each count type, report:

| Item | Actual | AGENTS.md | README.md | Skills.md | CLAUDE.md | Match? |
|------|--------|-----------|-----------|-----------|-----------|--------|

4. Check Cross-References

Verify that every [text](path) link in these files points to an existing file:

• AGENTS.md
• docs/WORKFLOW.md
• docs/QUALITY_SCORE.md
• docs/GOLDEN_PRINCIPLES.md
• CLAUDE.md
• Skills.md

Report any broken links.

5. Check Agent Table Completeness

Verify the agent table in AGENTS.md lists every .agent.md file found in Step 1. Report any agents that exist on disk but are missing from the table.

6. Output

Produce a summary with:

• PASS items where documented counts match actual counts
• FAIL items where counts are mismatched (with actual vs documented)
• WARN broken cross-reference links
• Recommended fixes (exact file, line, old value -> new value)

When to Run

• Before every release
• After adding or removing agents, skills, instructions, templates, or prompts
• As part of the review checklist