Pre-install reputation check

SKILL-CHECK

Check an agent skill’s canonical hash against the skillsissue.ai detonation database before you install or activate it.

↓ Download SKILL-CHECK View SKILL.md

Agent Skills format · Python standard library · macOS, Linux, and Windows · no candidate content upload

Hash locally

Reproduce the same versioned SHA-256 tree digest used by the skillsissue.ai ingestion pipeline.

Query privately

Download one small lookup shard using only the first byte of the digest. Skill contents never leave your machine.

Gate installation

Return the recorded verdict and dashboard link. Unknown, pending, suspicious, and malicious results require review.

Install

One package, multiple agents

Download and extract the ZIP into the skill directory used by your agent. The archive already contains the top-level skill-check/ folder.

Codex · personal

mkdir -p ~/.agents/skills
unzip skill-check.zip -d ~/.agents/skills

Available to Codex across your local projects.

Claude Code · personal

mkdir -p ~/.claude/skills
unzip skill-check.zip -d ~/.claude/skills

Available across Claude Code projects as /skill-check.

Project-scoped

unzip skill-check.zip -d .agents/skills
# or
unzip skill-check.zip -d .claude/skills

Commit the matching directory when a team should share the check.

Other compatible agents

unzip skill-check.zip -d <agent-skill-directory>

Any Agent Skills-compatible host that loads a folder containing SKILL.md can use the package.

Use

Check before installation

Ask the agent

Use $skill-check to check ./candidate-skill
before installing it.

In Claude Code, invoke /skill-check ./candidate-skill.

Run the checker directly

python3 skill-check/scripts/check_skill.py \
  --json ./candidate-skill

JSON output and stable exit codes make the checker suitable for installer scripts and hooks.

Interpretation

A match is evidence, not a guarantee

Benign: A recorded isolated run did not trigger the current rules. Review provenance and permissions before installing.
Pending or not found: The candidate has no completed published verdict. Treat it as unreviewed, not safe.
Suspicious: Review the findings and highlighted execution chain before deciding whether to proceed.
Malicious: Do not install or activate the skill unless you are performing explicitly authorized analysis in isolation.

SKILL-CHECK is a pre-install workflow, not an operating-system enforcement boundary. A supporting agent or installer must invoke it before copying or activating a candidate skill.