skill

ClawSkills

Practical AI agent Skills (agentskills.io) for project governance, bounded coding loops, research & session memory — shaped by real QC/production process habits.

EnglandTong/ClawSkillscommunitylistedunreviewed

Use skill

Set up

Skills are installed by placing their folder where your agent loads skills.

Download ZIP from GitHub Open the repository

The download comes from github.com/EnglandTong/ClawSkills itself, not from our server, so what you get is exactly what its author published.

Install instructions come from the README

Skills do not use an MCP config. Follow the steps in the README below.

README

synced Oct 23, 2026

ClawSkills

Practical AI agent Skills for project work: governance, coding loops, research, and session memory.

What. Six installable Skills. An agent reads SKILL.md and follows the process. They work in Claude Code, Cursor, and 35+ other products that speak the open Skill format — no per-platform rewrite.

Who. England Tong. I coded in university and never became a full-time engineer. Most of my working years were on factory floors and in quality control (品管): gates, evidence, acceptance, stop-loss. That is the habit these Skills borrow.

Why. Agents are good at moving. They are less good at stopping when the evidence is thin, the scope has drifted, or someone is about to rubber-stamp their own work. Shop-floor QC already solved that class of problem. These Skills apply the same discipline to agent-run projects.

Skills

Skill	Version	Description
`agent-loop-engineering`	2.2.0	Run an already-authorized coding goal in loops: keep state, fix what breaks, show real evidence, and stop before the same agent signs final acceptance.
`cms-project-governance`	2.2.1	Keep a drifting or half-finished project under control: one current delivery state, a right-sized scope, and independent QA acceptance.
`web-search-rules`	4.1.0	Search the web and save findings with a source URL, date, and quote on every claim.
`ai-workflow-os`	2.1.0	Pick the right Skill for a mixed request — lifecycle, governance, coding, memory, research — without creating a second source of truth.
`project-lifecycle-navigator`	2.1.1	Audit a project you are unsure about and get a go / narrow / pivot / archive / stop recommendation. Read-only: it does not write code or change governance state.
`daily-workflow`	4.1.0	Say 开工啦 or 收工啦 and get a resumable project note, so tomorrow you or another AI can pick up without re-reading everything.

Versions in this table match each Skill's SKILL.md, _meta.json, and (where present) skill.json. History lives in CHANGELOG.md.

Recommended Entry

Use cms-project-governance when a goal is vague or changing, old project records conflict, work needs sizing or alignment, or independent acceptance is required. Use agent-loop-engineering once the outcome, scope, acceptance, and authority are coherent.

cms-project-governance 2.2.1
  -> compact ACTIVE_PACKET (contract_version 2.0)
  -> agent-loop-engineering 2.2.0
  -> Ready for Independent Acceptance
  -> another agent, task, or human QA

Both Skills remain independently installable. Do not merge either ClawHub listing into the other.

For a combined workflow request, use ai-workflow-os as a router only. It delegates lifecycle advisory to project-lifecycle-navigator, session memory to daily-workflow, web research to web-search-rules, formal governance to cms-project-governance, and authorized coding to agent-loop-engineering. Each specialist stays independently installable and owns its own state.

Install Anywhere

SKILL.md is an open standard (agentskills.io) read by 35+ agent products, including Claude Code, ChatGPT, Codex CLI, VS Code, GitHub Copilot, Gemini CLI, Cursor, Trae, and Windsurf.

# from this repository (canonical source)
npx skills add EnglandTong/ClawSkills

# from ClawHub (one listing at a time)
clawhub install englandtong/daily-workflow

The ClawHub slug for cms-project-governance is coding-management-system (historical name; do not rename). Other Skills use the directory name as the slug.

Install one Skill, not all six. Most tools only need the one matching the job:

If you want	Install
Resumable session memory and handoff	`daily-workflow`
Web research saved with sources and dates	`web-search-rules`
Decide whether to continue, cut, or archive a project	`project-lifecycle-navigator`
Non-technical goal discovery, sizing, and QA acceptance	`cms-project-governance`
Autonomous coding loops under a clear authority boundary	`agent-loop-engineering`
One router across all of the above	`ai-workflow-os`

One rule. This GitHub repository is the canonical source; ClawHub and every cross-platform index are publish targets. Change order is always local, then CHANGELOG.md, then commit and push, then refresh the indexes. Never write a platform-specific content variant.

What the coding Skills actually enforce

These are current rules, not a frozen 2.1 release note. See CHANGELOG.md for the dated history.

Controller -> Developer -> QC prompts map into the loop. QC means Stage Reviewer inside a single-agent loop — not final sign-off.
Standard and Full execution cannot self-sign final acceptance.
Legacy Bootstrap indexes Docs/docs, reads selected current authority only, and writes nothing on conflict.
Compact context defaults to the Packet, current Work Order, affected source/tests, verification configuration, and last three Loop records.
Old JSONL gaps are aggregated and detailed findings are capped at 20 unless --strict-history is requested.
Runtime, Contract, Governance, Artifact, and Mixed delivery claims use distinct evidence rules. A passing unit test is not proof that the flow works.
Real paths are checked before project writes, so .., symlink, or junction escapes are rejected.

Local Commands

Same node invocations on bash and PowerShell:

node skills/agent-loop-engineering/scripts/bootstrap-active-packet.mjs --workspace <project> --language zh-CN --json
node skills/agent-loop-engineering/scripts/validate-loop-state.mjs --workspace <project> --json --summary --max-findings 20
node skills/agent-loop-engineering/scripts/test-state-tools.mjs

node skills/agent-loop-engineering/scripts/bootstrap-active-packet.mjs --workspace <project> --language zh-CN --json
node skills/agent-loop-engineering/scripts/validate-loop-state.mjs --workspace <project> --json --summary --max-findings 20
node skills/agent-loop-engineering/scripts/test-state-tools.mjs

Bootstrap is read-only unless --write is supplied and current authority is conflict-free.

Run before every release:

bash scripts/preflight.sh          # version sync, bundled copies, clean scan, frontmatter, quickstarts
bash scripts/publish-clawhub.sh    # dry run by default; add --yes to publish

preflight.sh exits non-zero on any failure. publish-clawhub.sh always passes --tags and --topics explicitly, because the ClawHub CLI defaults to latest and silently replaces the whole tag history when they are omitted.

Bundle Plugin

plugins/ai-engineering-expert version 1.1.3 bundles cms-project-governance 2.2.1 and agent-loop-engineering 2.2.0 as a ClawHub bundle-plugin and Qoder Expert Kit. It includes OpenClaw, Claude, and Qoder manifests plus bilingual routing instructions.

The bundled copies under plugins/ai-engineering-expert/skills/ must stay byte-identical to skills/. After any Skill change, re-copy and verify with diff -rq. A matching version number with divergent content breaks the release rule.

The Plugin is an extra one-install distribution format. Keep both standalone Skill listings live and searchable.

Release zips are not checked into this repository. Publish the ClawHub bundle-plugin from plugins/ai-engineering-expert or its exact committed GitHub source. A Qoder ZIP, if you build one, is not a ClawPack upload.

Repository Layout

plugins/
  ai-engineering-expert/
skills/
  agent-loop-engineering/
  cms-project-governance/
  ai-workflow-os/
  daily-workflow/
  project-lifecycle-navigator/
  web-search-rules/

License

Released under the MIT License. See LICENSE.