# Documentation Update

> Use when an implementation slice changed behavior and the canonical documentation must catch up before the slice is treated as done: updating the owning docs, updating manifests, and rebuilding the graph index when knowledge-graph nodes changed. This is the documentation_update action of the implementation stretch — one Agent-tool dispatch with subagent_type documentation-steward — distinct from the documentation_evidence_review action (review-evidence skill, which later verifies this work) and from the knowledge-maintenance actions (which repair repository-wide drift, not this slice's own docs).

- Skill: `bhangeef16/documentation-update` (Agent Skill)
- Install (CLI): `npx skillmds@latest add bhangeef16/documentation-update`
- Raw SKILL.md: https://api.skillmd.com/api/skills/bhangeef16/documentation-update/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: BhangeeF16 (https://skillmd.com/u/bhangeef16)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/bhangeef16/documentation-update

---


# Documentation Update

Update the canonical owners for what actually changed. Documentation debt created by a slice is paid inside the slice, not deferred to a someday cleanup.

## Dispatch

One synchronous `Agent`-tool call, `subagent_type: documentation-steward`.

## Required inputs

- The changed behavior — what the slice actually did, from its own evidence.

## Required outputs (all mandatory)

- Updated docs — the canonical owner for each changed fact, updated in place.
- Updated manifests, when a manifest points at anything the change touched.
- A graph reindex when knowledge-graph nodes changed.

## Rules — harvested documentation discipline

- If a fact's canonical owner cannot be determined, stop and ask — never invent a new owner or duplicate the fact across two documents.
- When the change moved or deleted a path, every active reference to it is repaired — not just noted — before this action reports pass.
- When the index genuinely cannot be refreshed (tooling unavailable, not merely "not run yet"), stop and record exactly why rather than silently treating the action as complete without it.

## Result

Report each required output produced (with its location), the result (pass / pass_with_risks / blocked / failure), and blocking questions as an explicit list (empty list stated explicitly).