SkillAgentSearch skills...

skill-governance

Governs the Claude Code/Codex Skill surface: reconciles source/install/catalog-visibility drift without losing cold capability

Install / Use

npx skills add daymade/claude-code-skills --skill skill-governance

Installs into whichever agent you are using.

About this skill
📄

SKILL.md

Installable skill definition

Quality Score

90/100

Supported Platforms

Claude Code
OpenAI Codex

Tags

Our assessment of skill-governance

skill-governance scores 90/100 on our quality scale, 1376th of 4,700 Development & Engineering skills we index (top 30%).

Its SKILL.md is 7.4 KB long, well organised into 9 sections with 3 code examples: a thorough specification that gives an agent plenty to work with.

With 1,426 GitHub stars, it is one of the more widely adopted skills in the catalogue.

Substance
29/30
Structure
18/20
Description
15/15
Adoption
13/20
Freshness
15/15

Maintenance, license and trust

  • The repository was last updated about 2 months ago, so skill-governance is actively maintained.
  • Our last check on 2026-10-30 found the source still online.
  • It is released under the MIT license, a permissive license that allows use, modification and commercial use with attribution.
  • Its trust signals score 100/100, with no cautions. These come from repository metadata, not a code audit — read the skill file before letting an agent act on it.

Safety scan

No issues found

Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful.

AI review by kimi-k2.7-code on 2026-10-03. Automated pattern scan on 2026-10-03. It catches known dangerous patterns, not every risk — read a skill before letting an agent act on it.

skill-governance compared with similar skills

All 4 of these similar skills score higher than skill-governance; compare them before choosing.

SkillScoreStarsUpdatedFormat
skill-governance (this skill)by daymade901.4k53d agoSKILL.md
taste-skillby Leonxlnx100116.8k8d agoCLAUDE.md
ai-job-searchby ai-job-search-dev10072.1ktodayCLAUDE.md
claude-howtoby luongnv8910047.3k2d agoCLAUDE.md
lobehubby lobehub10085.6ktodayMCP Server

Frequently asked questions

How do I install skill-governance?
Run npx skills add daymade/claude-code-skills --skill skill-governance. The install tabs above show the steps for each supported agent.
Which AI agents does skill-governance work with?
It is written for Claude Code and OpenAI Codex, as a SKILL.md file. Other agents that read the same format can often use it too.
Is skill-governance safe to use?
Our scan of the whole file found no instruction hijacking, hidden characters, credential access, data exfiltration or destructive commands. An AI review of the same text found nothing harmful. It is MIT-licensed and scores 100/100 on trust signals. Skills are instructions an agent will follow, so read the file before installing it and do not approve commands you do not understand.
Is skill-governance still maintained?
The repository was last updated about 2 months ago, so skill-governance is actively maintained.

name: skill-governance description: >- Governs the Claude Code/Codex Skill surface: reconciles source/install/catalog-visibility drift without losing cold capability. Use when users ask how many Skills load, why the catalog is truncated, want routers visible while other bundles stay cold, or need duplicate/superseded-Skill or old-cache cleanup. Not for building a marketplace manifest (use marketplace-dev) or repo audits (use marketplace-health-check).

Skill Governance

Govern the Skill surface the user actually experiences: the intended hot or router entries are fully visible in a fresh host, cold capabilities remain reachable, editable behavior has one canonical owner, and every retirement is recoverable. A smaller directory count is not the outcome.

Required outcome contract

Before acting, state in one sentence:

  • which entries or routers must be model-visible;
  • which capabilities must remain available cold;
  • which layer the user authorized changing;
  • what fresh-host evidence will prove success.

If the request is only “how many / what is loaded / why”, stay read-only.

System model

Never collapse these layers into one:

  1. Canonical source — where owned behavior may be edited.
  2. Installed inventory — bundles and versions that exist on disk.
  3. Discovery policy — what Claude or Codex may discover.
  4. Model-visible catalog — metadata in a fresh model prompt.
  5. Runtime resources — hidden scripts, references, and assets a router still needs.

Installed does not mean active; active does not prove visible; visible does not prove usable. Counts and byte totals are diagnostic values only.

Authority order

Use current runtime truth, not remembered conventions:

  • owned source repos and their manifests for editable Skill behavior;
  • claude plugin ... --json for current Claude marketplace/install state;
  • the explicit source-sync activation manifest for managed Daymade links;
  • ~/.agents/skills as Codex's current user Skill root;
  • exact-path ~/.codex/config.toml policy for Codex discovery disables;
  • codex debug prompt-input for the actual fresh Codex catalog;
  • the installed vendor bundle for third-party cold resources.

~/.codex/skills is legacy/system compatibility unless a current local contract explicitly assigns it another role. Never move third-party inventory into an owned-source activation manifest just to make ownership look complete.

Route the request

Read the named section of references/skill-surface-governance.md completely before using that workflow.

| Request | Read and use | |---|---| | What Codex really loads; count, truncation, duplicate identity, missing router | §3–4, then §11 | | Reconcile owned source links or ~/.agents/skills activation | §2–5, then §11 | | Keep gstack/Lark/IMA/UiPath or another bundle cold behind a router | §2–4, §6, then §11 | | Claude marketplace/plugin source or installed-state drift | §2–3, §7, then §11 | | Old cache versions | §2 and §7 “Exceptional manual cache repair” | | Standalone plugins superseded by a suite | §7–8, then §11 | | Project .claude/skills vs .agents/skills drift | §9, then §11 | | Retire loose or duplicate Skill directories | §2–3, §10–11 | | Decide which Skills stay model-visible vs user-invocable-only | §15, then §11 |

Fast read-only Codex audit

For a newly registered Skill whose local use is part of the delivery, run the target-name checks in references/skill-surface-governance.md §14 before calling it ready. This covers Claude's fresh command catalog and Codex's fresh prompt. The name must come from the requested delivery, not only the active whitelist.

Run from this Skill bundle:

python3 scripts/audit_codex_skill_surface.py --json

Only add policy the user or activation SSOT actually declared:

python3 scripts/audit_codex_skill_surface.py \
  --require-visible gstack-router \
  --json

The script compares codex debug prompt-input with the complete metadata parsed by Codex's own app-server skills/list, plus exact activation/discovery policy. Exit 0 is clean, 1 is pressure or drift to classify, and 2 means the evidence is invalid. It is read-only. Do not convert exit 1 into automatic pruning.

For a project's dual roots:

python3 scripts/audit_project_skill_roots.py <project-root> --json

That audit pairs direct child bundles by frontmatter name, recognizes only its explicit fail-visible compatibility-router contract, and distinguishes shared targets, identical copies, real drift, and invalid state.

Non-negotiable safety boundaries

  • Drift checks are read-only. Config edits, link sync, installs, uninstalls, moves, cache repair, and marketplace source changes require authorization.
  • Preserve Claude plugin scope. Verify replacements before retiring old identities.
  • Never use direct cache copying as installation or source sync.
  • Do not enforce “one cache version”. Current Claude Code owns orphan-version grace for running sessions; manual cache removal is exceptional repair only.
  • Do not use blind marketplace remove-then-add. Removing the last scoped marketplace can uninstall its plugins.
  • Read every candidate's unique instructions, scripts, references, and assets before calling it redundant. Old or short does not mean valueless.
  • Decide model-visible vs user-invocable-only per Skill from usage evidence (§15), never from its description or install location alone.
  • Keep cold third-party resources installed; hide only their exact discovery paths, then prove the router still resolves one representative capability.
  • Retire by recoverable move plus file/executable/hash manifest, never by rm -rf.
  • Existing sessions retain startup metadata. Restart before treating the interactive catalog as verification.

Source and activation ownership

For Daymade source-backed Codex activation, route to the current claude-switch-models-setup dry-run/apply workflow. Its explicit codex-active-skills.json owns only links created from declared source marketplaces. Do not reimplement its collision, symlink, or pruning logic here.

For Claude plugins, inspect current marketplace and install JSON, update or reinstall through the official CLI at the original scope, and independently read back the result. Treat cache folders as derived runtime artifacts.

For suite topology changes, use marketplace-dev to edit the source manifest; use this Skill only to reconcile already-landed migrations on the current host.

Definition of done

All applicable claims must be proven independently:

  • canonical source and current owner are named;
  • selected direct entries/routers appear in a fresh prompt under the requested discovery policy; report description truncation separately, including when the installation-only target gate passes;
  • entries intended cold are absent from that catalog;
  • one representative cold capability still resolves and works;
  • source-backed links or Claude installs read back with the intended identity, source/version, and scope;
  • any retired bundle and its recovery manifest still exist;
  • unresolved ownership, host-version behavior, or deliberately retained exceptions are explicit.

Use the reference's task-behavior check only when actual task execution is part of delivery; catalog visibility alone does not prove that result.

Stop there. Do not create a new hook, manifest, report layer, or cleanup project unless the requested outcome still lacks evidence.

Related Skills

View on GitHub
GitHub Stars1.4k
CategoryDevelopment
Updated1mo ago
Forks219

Languages

Python

Trust signals

100/100

From repository metadata: license, adoption, age and documentation. Not a code audit — see the Safety scan above for what the skill file itself contains.

No cautions