Docs Writer

In Scope

All markdown documentation except agent-output/**/*.md:

• docs/ — user-facing docs (quickstart, workflow, troubleshooting, etc.)
• docs/prompt-guide/ — agent & skill prompt examples
• docs/exec-plans/tech-debt-tracker.md — tech debt inventory
• README.md — repo root README
• CONTRIBUTING.md — contribution guidelines
• CHANGELOG.md — release history
• QUALITY_SCORE.md — project health grades
• .github/instructions/docs.instructions.md — architecture tables

Out of Scope (Has Own Validators)

Step-by-Step Workflows

Workflow 1: Update Existing Documentation

1. Identify target files: Determine which files in docs/ need updates.
2. Read latest version: Always read the current file before editing.
3. Load standards: Read references/doc-standards.md for conventions.
4. Apply changes: Follow the doc-standards conventions strictly:
   • 120-char line limit (CI enforced)
   • Single H1 rule (title only)
   • File header: # {Title} + > Version {X.Y.Z} | {description}
   • Version number from VERSION.md (single source of truth)
5. Verify links: Check all relative links resolve to existing files.
6. Run validation: Offer to run npm run lint:md and npm run lint:links.

Workflow 2: Add Documentation for New Entity

When a new agent or skill is added to the repo:

1. Read architecture: Load references/repo-architecture.md for current entity inventory and naming conventions.
2. Identify all files needing updates:
   • New agent → update docs/README.md agent tables, .github/instructions/docs.instructions.md agent count/table
   • New skill → update docs/README.md skill tables, .github/instructions/docs.instructions.md skill count/table
3. Match existing patterns: Study adjacent entries in each table to match column format, emoji conventions, and description style.
4. Update counts: Increment totals in section headings (e.g., "## Skills (10)" → "## Skills (11)").
5. Cross-reference check: Search for other files mentioning the old count and update them too.

Workflow 3: Freshness Audit (Staleness Check)

1. Load checklist: Read references/freshness-checklist.md.
2. Scan each audit target:
   • Version numbers match VERSION.md
   • Agent/skill counts match filesystem
   • Tables list all entities present in filesystem
   • No references to removed/renamed agents
3. Check project health files:
   • Read QUALITY_SCORE.md — verify grades still reflect reality
   • Read docs/exec-plans/tech-debt-tracker.md — verify items still relevant
4. Report findings: Present a table of issues found with:
   • File path, line number, issue description, suggested fix
5. Auto-fix: For each issue, propose the exact edit and apply it after user confirmation (or immediately if user said "fix all").
6. Update health metrics: If fixes change quality grades, update QUALITY_SCORE.md.

Workflow 4: Explain the Repo Architecture

1. Load architecture: Read references/repo-architecture.md.
2. Answer questions: Use the reference to explain how components connect — agents, skills, instructions, templates, artifacts, and the 7-step workflow.
3. Cite sources: Point to specific files when answering.
4. Stay current: If the reference seems outdated vs. filesystem, note the discrepancy and offer to update the reference.

Workflow 5: Generate Changelog Entry

1. Find last version tag: Run git tag --sort=-v:refname | head -1.

2. Get commits since tag: Run git log --oneline {tag}..HEAD --no-merges.

3. Classify by type: Map conventional commit prefixes to Keep a Changelog sections:

   • feat: → ### Added
   • fix: → ### Fixed
   • docs:, style:, refactor:, perf:, test:, build:, ci:, chore: → ### Changed
   • feat!: or BREAKING CHANGE: → ### ⚠️ Breaking Changes

4. Format entry: Match the style in CHANGELOG.md:

   ## [{next-version}] - {YYYY-MM-DD}
   
   ### Added
   
   - Description of feature ([commit-hash])
   
   ### Changed
   
   - Description of change ([commit-hash])
   
   ### Fixed
   
   - Description of fix ([commit-hash])
   

5. Determine version bump:

   • Breaking change → major
   • feat: → minor
   • fix: only → patch

6. Present to user: Show the formatted entry for review before inserting into CHANGELOG.md.

Workflow 6: Proofread Documentation

A three-layer review: language quality, tone/terminology, and technical accuracy.

1. Select scope: Ask user which files to review, or default to all files in docs/.

2. Layer 1 — Language quality:

   • Run npm run lint:prose (Vale) for automated prose checks.
   • Manually scan for: grammar errors, spelling mistakes, passive voice, awkward phrasing, overly long sentences (>30 words).

3. Layer 2 — Tone and terminology:

   • Verify consistent terminology against docs/GLOSSARY.md.
   • Check tone is active and action-oriented (not academic/passive).
   • Flag jargon not defined in the glossary.
   • Ensure agent/skill names use exact casing from their frontmatter (name: field) — e.g., "Bicep Code" not "bicep code agent".

4. Layer 3 — Technical accuracy:

   • Load references/repo-architecture.md for ground truth.
   • Verify agent/skill counts, names, and descriptions match the actual filesystem.
   • Confirm workflow step numbers and artifact filenames are correct.
   • Check that capability claims are truthful (e.g., if a doc says "supports 8 skills", verify 8 skill folders exist).
   • Cross-check version numbers against VERSION.md.

5. Report findings: Present a table per file:

   | #   | Line | Layer       | Issue                      | Suggestion       |
   | --- | ---- | ----------- | -------------------------- | ---------------- |
   | 1   | 12   | Language    | Passive voice              | Rewrite actively |
   | 2   | 34   | Terminology | "IaC tool" not in glossary | Use "Bicep"      |
   | 3   | 56   | Accuracy    | Says 6 agents, actual is 8 | Update count     |
   

6. Apply fixes: After user review, apply corrections. For language/tone fixes, show before/after for each change. For accuracy fixes, apply directly (same as freshness audit).

Workflow 7: Process Freshness Issues

Trigger: "Fix the docs freshness issue" or auto-created GitHub issue with docs-freshness label

1. Read the issue body for the findings table
2. For each finding, apply the appropriate fix from the freshness checklist
3. Run npm run lint:docs-freshness to verify 0 findings remain
4. Summarize changes made

Guardrails

• Never modify files in agent-output/, .github/agents/, or .github/skills/azure-artifacts/templates/
• Always read the latest file version before editing
• Always verify line length ≤ 120 characters after edits
• Preserve existing Mermaid diagram theme directives
• Use VERSION.md as the single source of truth for version numbers

Troubleshooting

References

• references/repo-architecture.md — Repo structure, entity inventory
• references/doc-standards.md — Formatting conventions, validation
• references/freshness-checklist.md — Audit targets and auto-fix rules

Reference Index