agentleFS
kaisdaviskaisdavis/claude-skills/docs-update/SKILL.md

docs-update

Skill1 starsChanged 6 months agoMITView on GitHub

Use when finishing a session in a repo with a multi-file doc set (any 3+ of README.md, CLAUDE.md, CONTEXT.md, ARCHITECTURE.md, DECISIONS.md, WORKLOG.md, TODO.md) and the docs are behind the code. Routes each delta to the right file: appends to WORKLOG, edits TODO, records decisions, adds CONTEXT terms, updates ARCHITECTURE. Triggers on the user saying "catch up the docs", "sync the docs", "update the docs", "doc catch-up", "bring the docs current", "the docs are stale", "what about docs", "wrap up", "log this work", or invoking `/docs-update`. Pairs with the `context-md` skill (the CONTEXT.md-specific worker docs-update delegates to for glossary edits). Skip in single-doc repos (README only) and in repos with no doc set yet (use `context-md` first).

  • Commits and pushes

17 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 5 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 16 edited versions we read keep these.

  • 5

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    After a one-line fix that doesn't shift docs (typo, lint, dependency bump).

    line 21
  2. Distinctive:

    Don't guess shape from filename.

    line 35
  3. Distinctive:

    Use the in-context session arc (what happened in chat) as primary source of intent.

    line 47
  4. Distinctive:

    Before writing, show the user a one-line summary per doc:

    line 65
  5. Distinctive:

    Do NOT push without explicit approval.

    line 91

“Common” is measured: kept by at least 6 in 10 of the 16 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

docs-update

Catch up the per-repo doc set with what actually happened this session. Read commits, code delta, worklog, and the chat arc; update each doc by appending the right kind of change to the right file.

The whole point: every doc has a job. Don't pile updates into whichever doc is closest. Route the right change to the right file.

When to use

  • End of a work session in a repo with 3+ of the canonical docs above.
  • The user says "catch up the docs", "sync docs", "doc catch-up", or invokes /docs-update.

When NOT to use

  • Repo has only README.md (single-doc repo). Just update the README in place; no skill needed.
  • Mid-session refactor without a meaningful arc yet. Wait for the work to land.
  • After a one-line fix that doesn't shift docs (typo, lint, dependency bump).

Process

1. Inventory

ls *.md | grep -E '^(README|CLAUDE|CONTEXT|ARCHITECTURE|DECISIONS|WORKLOG|TODO)\.md$'

Note which of the 7 are present. Other md files (CHANGELOG, CONTRIBUTING, etc.) are out of scope for this skill.

2. Read each one

Don't guess shape from filename. Read the actual file. Note its current structure and which section is the append target (e.g. WORKLOG has a today-dated section; TODO has Open and Recently shipped).

3. Survey session changes

git log --oneline <since-marker>..HEAD       # commits this session
git diff --stat <since-marker>..HEAD          # files touched
git status --short                            # uncommitted

For <since-marker>: prior session's last commit, or today's first commit. Fall back to "last 24h" if unsure (--since='24 hours ago').

Use the in-context session arc (what happened in chat) as primary source of intent. The git log gives the artifacts.

4. Per-doc decision rules

For each doc that exists, decide: append, edit, or skip.

FileAppend/edit when
WORKLOG.mdAlways (if any code shipped). One paragraph under today's date heading. New ## YYYY-MM-DD section if no entry for today.
TODO.mdWhen new open items surfaced OR existing items were shipped. Strike with ~~~~ OR move to "Recently shipped" section. Add new items to "Open".
DECISIONS.mdOnly when a new design decision passes the 3-gate: hard to reverse, surprising without context, real trade-off. Otherwise skip. Most commits don't warrant a new decision entry.
CONTEXT.mdWhen new domain vocabulary was introduced. New noun in the system, new verb in the workflow, new flagged ambiguity. NOT for general programming concepts. (Delegate to the context-md skill.)
ARCHITECTURE.mdWhen module map changed (new file in src/), phase count changed, new external integration added, or storage shape changed.
README.mdWhen operator-facing surface changed: new CLI flag, new env var, new install step, new run command. Architecture section update only if ARCHITECTURE.md changed and you want them mirrored.
CLAUDE.mdWhen a new convention was established that future-session-Claude needs to know. Rare. Most session learnings belong in a memory system, not CLAUDE.md.

5. Preview

Before writing, show the user a one-line summary per doc:

docs-update preview:
  WORKLOG.md:     +1 entry under 2026-05-17 (slice B, link-mode, duration fix)
  TODO.md:        +1 open (non-Shorts handling), 0 struck
  DECISIONS.md:   no change
  CONTEXT.md:     +2 terms (Link-mode, Persona facts)
  ARCHITECTURE.md: no change
  README.md:      no change
  CLAUDE.md:      no change

proceed? (y / edit which one / skip)

On y, write each one. On edit X, ask the user for the correction. On skip, abort.

6. Commit

One commit covering all doc changes:

git add WORKLOG.md TODO.md CONTEXT.md   # etc., only changed files
git commit -m "docs: catch up after <one-line session topic>"

Do NOT push without explicit approval. This skill commits; pushing (especially to a deploy branch that auto-builds) is a separate decision the user makes.

Common mistakes

  • Padding WORKLOG with trivia. One paragraph per session, summarizing the arc. The git log has the detail. Don't restate every commit message in prose.
  • Decision-record inflation. "We chose X over Y" is a decision IF Y was a real alternative and the cost of reversing is real. Most refactors don't warrant a record.
  • General programming terms in CONTEXT.md. "Timeout" is not domain vocabulary. "Bundle" (when your system has a specific Bundle concept) is.
  • Duplicating content across files. WORKLOG is the diary, DECISIONS is the why, ARCHITECTURE is the shape, CONTEXT is the vocabulary, README is operator-facing, CLAUDE is agent-facing. They reinforce, they don't repeat.
  • Pushing without approval. git push to a deploy branch is a separate ask. This skill commits; deploy-decisions are the user's.

Red flags (stop and reconsider)

  • About to write more than one paragraph to WORKLOG. Probably padding. Trim.
  • About to add a 5-bullet decision record. Probably belongs as a single line in WORKLOG or a single bullet in TODO instead.
  • About to mention "Claude" or "agent" in WORKLOG. The diary is for humans; agent learnings go to a memory system.
  • About to add CLAUDE.md entries that overlap with an existing memory entry. Memory is the right home for cross-repo learnings.

Quick reference

docs-update flow:
  inventory  →  read each  →  survey delta  →  per-doc decide
                                                      ↓
                                                  preview
                                                      ↓
                                                  user: y
                                                      ↓
                                                  write + commit
                                                  (no push)

Related:

  • The context-md skill: CONTEXT.md format and the glossary-maintenance worker docs-update delegates to.
  • Continuous-diary cadence: append WORKLOG entries as work happens, not batched at session end.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.