← Back to Skills Marketplace
jzg-lab

Docs Refresh

by jzg-lab · GitHub ↗ · v1.0.0 · MIT-0
cross-platform ✓ Security Clean
108
Downloads
1
Stars
0
Active Installs
1
Versions
Install in OpenClaw
/install docs-refresh
Description
Refresh authoritative docs through a routed progressive-disclosure workflow.
README (SKILL.md)

Docs Refresh

Use this to close out work by syncing durable docs with the repository's real state. It is not a general writing prompt.

Keep the workflow platform-neutral. Host-specific aliases, launcher syntax, and metadata belong in adapters.

Shared Rules

  • The repository is the record system. Durable knowledge belongs in versioned docs near the code, not in chat.
  • AGENTS.md is a map, not a manual. Keep it short, stable, and navigational.
  • Update the smallest authoritative doc that owns the fact.
  • Prefer the bundled collector before manual routing, but do not block on it if it cannot be run from the skill directory.
  • Never run git add, git commit, or automatic VCS cleanup.

Route First

  1. Read repository guidance first. Start with AGENTS.md when present, then follow only the pointers needed to find authority, ownership, freshness rules, and doc-lint requirements.
  2. Resolve the directory containing this SKILL.md.
  3. If you can run bundled files from the skill directory, run the collector from that skill directory against [repo-root]. Do not assume the target repo has its own scripts/collect_changed_context.sh.
  4. If the collector succeeds, read its [routing] section and open only the matching mode file from preferred_mode_doc.
  5. If the collector is unavailable, cannot be resolved from the skill directory, or shell execution is unavailable, manually collect git status --short, staged and unstaged changed files, untracked files, diff stat, and the repository docs layout.
  6. Use the manual routing fallback below to choose exactly one mode file.
  7. Use the current workspace state as the source of truth: staged diff, unstaged diff, untracked files, and generated artifacts when the repository treats them as authoritative outputs.
  8. Inspect only the code, schema, generated artifacts, and current docs needed to confirm the real behavior change.
  9. Stop after explaining what changed or why no doc change was needed.

Manual Routing Fallback

Use this only when the bundled collector cannot be run from the skill directory or its routing output is unavailable.

  • bootstrap: no split docs tree exists, and the repository does not yet have a stable docs system beyond README.md or a few incidental docs. README.md only is still bootstrap.
  • minimal: no split docs tree exists, but the repository already has one or more stable living docs such as AGENTS.md, ARCHITECTURE.md, or durable top-level current-state docs. A repo can be minimal even if AGENTS.md is absent.
  • structured: split docs domains already exist, such as docs/design-docs/, docs/product-specs/, docs/references/, docs/generated/, or docs/exec-plans/.
  • repair: a real docs system exists, but its map or authority is broken enough that content edits would deepen drift: missing index pages, stale or broken AGENTS.md or ARCHITECTURE.md, broken cross-links, or no usable navigation path from entry docs to owning pages.
  • Missing AGENTS.md alone does not mean repair.
  • Sparse docs alone do not mean repair.
  • When in doubt between minimal and repair, choose minimal unless the repository already has a real docs system whose map is broken.

Shared Decision Rules

  • Usually no documentation edits for tests-only, docs-only, comments-only, formatting-only, or internal refactors that do not change public behavior, architecture, state semantics, runtime operation, or external contracts.
  • Force a documentation review when the change touches APIs, schema, CLI flags, run modes, scheduler behavior, trigger flow, core architecture, state model, external contracts, durable entry points, documented core beliefs, or stale navigation.
  • If the signal is ambiguous, inspect the diff before deciding. Do not update docs just because code changed.

Shared Authority Rules

  • Prefer this authority order unless the repository defines a stricter one:
    1. code, tests that prove behavior, schema, and authoritative generated artifacts
    2. current-state architecture, design, product, reliability, security, and reference docs
    3. navigation docs and indexes such as AGENTS.md, ARCHITECTURE.md, and docs/**/index.md
    4. plans, decision logs, and history documents
  • AGENTS.md never outranks code or the deeper docs it points to.

Modes

  • modes/bootstrap.md: Repos with no real docs system yet. Start with phased growth from README.md, AGENTS.md, and only add deeper structure when durable domains appear.
  • modes/minimal.md: Repos with core living docs but no split docs taxonomy. Prefer updating those docs and only create the first subtree when it has earned a second durable page.
  • modes/structured.md: Repos that already have split docs domains. Preserve the existing taxonomy and converge to the smallest authoritative page.
  • modes/repair.md: Repos whose docs system exists but navigation or authority surfaces are stale. Repair the map and indexes before expanding content.
  • Each mode file supports both collector-selected routing and the manual fallback rules above.

Explain The Result

If no documentation change is needed, say so explicitly and explain why.

If documentation changed, report:

  • which documents changed
  • why those documents were the right convergence points
  • whether AGENTS.md stayed unchanged on purpose, or why its navigation changed
  • why you reused existing docs instead of creating new files
  • what remains stale, uncovered, or needs manual follow-up

Stop after the explanation. Do not stage or commit anything.

Usage Guidance
This skill appears to do exactly what it says: it analyzes a git repo's changed files and routes to a docs-update mode. Before allowing it to run, verify you are comfortable with the agent executing the bundled shell script and git commands against your repository (the collector uses bash and git). Review scripts/collect_changed_context.sh yourself (it only runs git status/diff/ls-files and inspects filenames/content patterns) and confirm there are no network calls or secret reads you didn't expect (none are present). If you do not want the agent to execute shell commands, run the collector locally in a sandbox or follow the SKILL.md's manual fallback steps instead. Finally, note the registry metadata omitted declaring 'git' (and a shell) as required binaries — this is a minor documentation gap but not a functional contradiction.
Capability Analysis
Type: OpenClaw Skill Name: docs-refresh Version: 1.0.0 The docs-refresh skill bundle is a well-structured tool designed to help an AI agent synchronize repository documentation with code changes. It uses a shell script (scripts/collect_changed_context.sh) to gather repository metadata via standard git commands and provides clear, non-destructive instructions in SKILL.md and various mode files (modes/*.md). The bundle lacks any indicators of malicious intent, such as data exfiltration, unauthorized network access, or persistence mechanisms.
Capability Assessment
Purpose & Capability
The skill's name, description, SKILL.md, and included scripts all align: its purpose is to inspect a git repository's changed files and route to a docs-update mode. Minor mismatch: the bundled collector script requires git and a POSIX shell (bash), but the registry metadata declares no required binaries. This is a small documentation inconsistency, not functional misalignment.
Instruction Scope
SKILL.md limits actions to reading repository state (git status/diff/ls-files), using the bundled collector if runnable, selecting a single docs mode, and explaining results. It explicitly forbids staging/committing. The instructions do ask the agent to read repo files and diffs (expected for this purpose) but do not direct data to external endpoints or require unrelated system access.
Install Mechanism
No install spec is present (instruction-only with bundled scripts). The code is included in the skill bundle; nothing is downloaded or executed from external URLs during install. Risk from install-time behavior is low.
Credentials
No environment variables, credentials, or config paths are requested. The scripts do not read secrets and only use git and filesystem traversal to classify changed files — proportional to the stated task.
Persistence & Privilege
The skill is not always-enabled and does not request persistent elevated privileges or modify other skills or global agent settings. It explicitly instructs not to stage/commit changes.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install docs-refresh
  3. After installation, invoke the skill by name or use /docs-refresh
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.0.0
docs-refresh is a portable documentation-refresh workflow packaged in a reusable prompt/skill layout. It treats the repository as the record system, starts from AGENTS.md as a map, inspects current workspace state, and uses progressive disclosure to load only the workflow branch that matches the repository's current docs maturity and foundation maturity.
Metadata
Slug docs-refresh
Version 1.0.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 1
Frequently Asked Questions

What is Docs Refresh?

Refresh authoritative docs through a routed progressive-disclosure workflow. It is an AI Agent Skill for Claude Code / OpenClaw, with 108 downloads so far.

How do I install Docs Refresh?

Run "/install docs-refresh" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is Docs Refresh free?

Yes, Docs Refresh is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does Docs Refresh support?

Docs Refresh is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created Docs Refresh?

It is built and maintained by jzg-lab (@jzg-lab); the current version is v1.0.0.

💬 Comments