agentsclimarketplace
Sign in

Skill guardian

Skill Adenine-AGCT/skill-guardian/skills/skill-guardian

Audit local agent skill directories, explain current version provenance, compare known GitHub upstreams, score trust and update risk, and finish with a clear recommendation about whether any local skills should be updated now.From its SKILL.md

Install
npx -y skills add Adenine-AGCT/skill-guardian --skill skill-guardian

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its file declares

Copied from the file, not written here

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

2.4 KB, 449 tokens by cl100k_base, as published. Nobody here has run it

Skill Guardian Skill

Overview

Use this skill when the user wants a dedicated skill for auditing installed Agent Skills before making update decisions.

Skill Guardian is built for skill audit and update governance. It starts with a quick, low-overhead audit, checks provenance and local drift first, and only escalates to deeper update checks when the evidence justifies it.

This skill is advisory. It does not auto-update installed skills.

Workflow

Run the bundled wrapper:

python .\scripts\skill_guardian.py

Use offline mode when network access is unavailable:

python .\scripts\skill_guardian.py --offline

The wrapper performs:

  1. Multi-root discovery for local skill directories
  2. A quick local structural and behavior audit with baseline and drift detection
  3. Lightweight upstream metadata checks to estimate version gaps
  4. Escalation to deeper remote comparison only when risk, drift, or version gap signals justify it
  5. Trust scoring and update recommendation
  6. A final human-readable summary of what should or should not be updated

Output Contract

The final response should contain:

  • how many local skills were found
  • which roots were scanned
  • which skills have recognized upstream provenance
  • which skills show local drift
  • which skills were escalated because of large version gaps
  • which skills are already current
  • which skills have updates available
  • which updates need review or should be blocked
  • whether the user should update anything now

Every per-skill summary should include:

  • source_type
  • version_status
  • baseline_status
  • version_gap_level
  • trust_score
  • risk_level
  • update_recommendation
  • confidence_explainer

Safety Rules

  • Never auto-update a skill
  • Never execute remote scripts while inspecting them
  • Prefer the quick audit path unless the evidence clearly justifies a deeper check
  • Treat changes under scripts/ or executable file additions as review-heavy
  • Treat unknown-source skills as auditable but not strongly version-resolvable
  • If remote inspection fails, still complete the local audit and say that the remote version check was unavailable

What ships with it: 5 files

86.5 KB alongside SKILL.md, 2 of them executable

agents/

assets/

references/

scripts/

Keep looking

Skills are one crate of 326,401. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.