Doc Sync

Documentation-Code Sync

Detect and resolve drift between code and documentation without front-loading a full repository audit. Treat standard Sonnet context as the baseline.

Invocation

/doc-sync <direction> [options]

Directions:

• code-to-docs: code is source of truth
• docs-to-code: docs are source of truth
• full: bidirectional reconciliation; conflicts require human review

Options:

• --scope "path": limit the run to a subtree or file group
• --incremental: inspect changed files first
• --dry-run: report only
• --parallel N: cap concurrent auditors; default 2, maximum 4
• --guidance "text": use operator guidance for ambiguous cases
• --no-commit: do not commit

Artifact root resolution

Treat every .aiwg/... path below as a logical artifact path. Before reading or writing sync state, working evidence, or reports, run aiwg artifacts path from the active workspace and use the returned absolute directory as AIWG_ARTIFACT_ROOT. Never write these payloads to a literal project-local .aiwg/ when AIWG_ARTIFACTS_PATH or .aiwg-location redirects the corpus.

Context Budget Rules

• Start with git status --short and git diff --name-only; do not read broad diffs before the changed-file set is known.
• If neither --scope nor --incremental is supplied, derive a candidate scope from changed files and ask before expanding to full repo.
• Prefer shell inventories (find, rg --files, git diff --name-only) over reading file bodies.
• Dispatch subagents only after scope is known. Use at most min(--parallel, 4) auditors, and prefer 2 for broad or release requests.
• Each auditor must return at most 10 findings and 600 words. It should write detailed evidence to $AIWG_ARTIFACT_ROOT/working/doc-sync/ and return only the path plus a summary.
• Do not preload other skills into subagents. If another capability is needed, invoke it after this skill finishes or in an isolated follow-up.

Workflow

1. Parse direction and options.
2. Build a cheap scope inventory:
   • git status --short
   • git diff --name-only
   • git diff --cached --name-only
   • rg --files docs README.md CHANGELOG.md package.json src agentic tools only when the changed-file set is insufficient.
3. Choose audit lanes from the scoped files:
   • CLI/API docs
   • provider/deployment docs
   • skill/agent catalogs
   • README, changelog, release, or blog material
   • config/schema docs
4. Run bounded auditors for only the selected lanes. Each prompt must include:
   • exact paths to inspect
   • direction
   • max 10 findings
   • max 600-word return summary
   • instruction to store detailed notes under $AIWG_ARTIFACT_ROOT/working/doc-sync/
5. Merge summaries into $AIWG_ARTIFACT_ROOT/reports/doc-sync-audit-{date}.md with:
   • scope
   • findings by severity
   • auto-fixable vs human-required items
   • files changed or proposed changes
6. If not --dry-run, apply high-confidence fixes only.
7. Validate modified files with targeted checks:
   • markdown link/anchor checks where available
   • npm run lint:claude-context if Claude-facing skills or agents changed
   • project-specific build/test checks only when source changed
8. Record $AIWG_ARTIFACT_ROOT/.last-doc-sync.
9. Commit only when requested by the surrounding workflow and not blocked by --no-commit; otherwise leave a concise final summary.

Release or Blog Coverage Requests

For combined requests such as "docSync code2doc, ensure the monthly blog covers the work, then commit-and-push":

1. Keep this skill to documentation drift detection and fixes.
2. Treat blog/release coverage as one selected audit lane with changed files plus month-bounded git history, not a full repository read.
3. Return a handoff summary for commit-and-push instead of invoking it inside this skill.

Output Contract

Return:

• audit report path
• files changed
• remaining human-review items
• validation commands run
• next recommended skill, if any