docs-sync

---

name: docs-sync description: Documentation synchronization principles — 3-layer model (public docs, commands, skills+agents), behavior-over-counts philosophy, staleness detection, and verification checklists. Rules are out of scope per ADR-032 Amendment 2026-04-26 (hand-curated via ADR, surfaced as read-only NOTEs). Use this skill whenever updating documentation after a commit that adds features, changes behavior, or modifies structure; whenever running /doks or /chekpoint with docs in the diff; when the user says "update docs", "sync docs", "docs are stale", "drift", "counts don't match", or "descriptions are outdated"; and after adding/removing agents, skills, commands, or hooks. The skill enforces description-over-count discipline: matching counts is the easy part, but descriptions drift silently and mislead future sessions — that's what this skill catches.

Documentation Synchronization

Principles for keeping all documentation layers accurate and in sync with code behavior. This skill teaches the methodology; doks agent executes the workflow.

When to Use

• After commits that add features, change behavior, or modify structure
• When any agent needs to update documentation
• During /doks or /chekpoint workflows
• When reviewing documentation accuracy

The 3-Layer Documentation Model

Every behavioral change must be reflected across all relevant layers. Rules (.claude/rules/) are explicitly out of scope — they are hand-curated operational logic updated via deliberate ADR, not via diff-driven sync (Amendment 2026-04-26 to ADR-032; rationale in docs/research/research-008-auto-loaded-rules-vs-on-demand-skills.md).

Layer	Files	What to Check
1. Public Docs	CLAUDE.md, README.md	Component counts, feature descriptions, status line
2. Commands	.claude/commands/*.md	Workflow steps, agent chains, skill references
3. Skills + Agents	.claude/skills//SKILL.md, .claude/agents/.md	Integration sections, hook references, API descriptions
Rules (read-only)	.claude/rules/**/*.md	NEVER edited by /doks. If a behavioral change implies a rule update, surface as a NOTE in output and stop.

Core Principle: Behavior Over Counts

Counts are easy. Descriptions are hard. Prioritize descriptions.

A hook that changed behavior but kept the same name needs a description update, not just a count check. Examples:

• Hook went from "logs tool results" to "logs tool results AND captures error messages" -- the description must change even though the hook count didn't
• Agent gained a new skill -- the catalog table needs the skill name, not just "+1"
• Command workflow added a new step -- the step must be documented, not just the step count

Ground Truth Extraction

Never trust memory or cached counts. Always verify from the filesystem:

ls .claude/agents/*.md | wc -l       # agent count
ls .claude/skills/*.md | wc -l       # skill count
ls .claude/commands/*.md | wc -l     # command count
ls .claude/hooks/scripts/*.js | wc -l # hook script count
ls .claude/rules/**/*.md | wc -l     # rule count
npx vitest run 2>&1 | tail -5        # test count

Staleness Detection

Finding What Changed

git log --oneline -10          # recent commits
git diff HEAD~N --stat         # changed files
git diff HEAD~N -- .claude/    # harness changes

For each changed file, answer:

1. What did it do BEFORE?
2. What does it do NOW?
3. Is this behavioral change documented anywhere?

Cross-Reference Check

# Find stale references in writable layers
grep -rn "hook_name" .claude/skills/ .claude/agents/ .claude/commands/
# Read-only check: surface a NOTE if rules mention the changed component
grep -rn "hook_name" .claude/rules/
# Compare with actual hook code
cat .claude/hooks/scripts/hook_name.js

Verification Checklist

After making documentation updates, verify ALL of these:

• Feature coverage -- each behavioral change documented in at least 2 layers
• Stale references -- no documentation references deleted/renamed components
• Rules surface check (read-only) -- if a rule mentions a changed component, append a NOTE to output for manual ADR follow-up. Never edit rule files.
• Skills sync -- skills referencing hooks/APIs describe current behavior
• Counts match -- documented counts equal filesystem counts
• Status line -- version number reflects the nature of the change

Anti-Patterns

Anti-Pattern	Why It's Wrong
Updating counts but not descriptions	Counts tell you how many; descriptions tell you what they do
Only updating CLAUDE.md	Rules and skills are loaded every session too
Trusting memory for current state	Files may have changed since you last read them
Stopping after Layer 1	Layers 2-3 are where Claude actually reads the documentation
Editing .claude/rules/ from /doks	Rules are out of scope (Amendment 2026-04-26). Auto-edit causes silent drift; updates happen only via deliberate ADR.
Documenting "planned" features as if they exist	Misleads future sessions into assuming capabilities

Language Preservation

• English files stay English (CLAUDE.md, README.md, rules, skills, commands)
• Spanish files stay Spanish (if any)
• Never mix languages within a single file

Integration

• Agent: doks (primary executor of this methodology)
• Command: /doks (triggers doks with this skill)
• Also useful for: any agent that needs to update documentation as a side effect

no_context Application

Generate documentation from code, not from assumptions. Read the actual source before writing descriptions. If a component is referenced but cannot be found on disk, flag it and remove the reference rather than guessing what it might do.