Back to skill
Skillv1.0.0
ClawScan security
git-commit-skill · ClawHub's context-aware review of the artifact, metadata, and declared behavior.
Scanner verdict
ReviewMar 15, 2026, 12:40 PM
- Verdict
- Review
- Confidence
- medium
- Model
- gpt-5-mini
- Summary
- The skill's behavior matches a Git commit helper and is mostly coherent, but a prompt-injection signal (unicode control characters) was detected in the SKILL.md which warrants caution before installing.
- Guidance
- This skill appears to do what it says (analyzes git diffs, suggests Gitmoji commit messages, and stages/commits/pushes with confirmations). However, a prompt-injection signal (hidden/unicode control characters) was found in the SKILL.md. Before installing, review the raw SKILL.md for invisible characters or unexpected content, and preferably run the skill in a test repository first. Ensure it always asks for your explicit Yes/No before git add/commit/push and do not grant it any additional credentials or persistent privileges. If you see zero-width or control characters in the file, remove them or request a clean copy from the publisher; treat that finding as suspicious until resolved.
- Findings
[unicode-control-chars] unexpected: The pre-scan detected unicode control / zero-width characters in SKILL.md. This is not expected for a simple instruction doc and could be used to obfuscate malicious instructions or to attempt prompt-injection. It should be manually inspected and cleaned; presence of these characters reduces trust in the file.
Review Dimensions
- Purpose & Capability
- okName/description (git commit generator using Gitmoji) aligns with the instructions: reading git status/diff, generating commit messages, staging/committing/pushing. No unrelated binaries, env vars, or external services are requested.
- Instruction Scope
- noteRuntime instructions limit actions to git commands (status, diff, add, commit, push) and include a clear multi-step confirmation flow and a mandatory sensitive-data check. This is appropriate for the stated purpose. However, the SKILL.md contains a prompt-injection detection (unicode control characters) which may be an attempt to manipulate agent behaviour or evaluations; the file should be inspected for hidden/zero-width characters before trusting automatic processing.
- Install Mechanism
- okInstruction-only skill with no install spec and no code to write to disk — minimal install risk.
- Credentials
- okNo environment variables, credentials, or config paths are requested. The sensitive-file and diff checks are reasonable and align with the commit helper role.
- Persistence & Privilege
- okalways: false and no indication the skill attempts to persist beyond normal runtime. It does stage/commit/push only after explicit user confirmations per the instructions.