Agent Safety

PassAudited by ClawScan on May 1, 2026.

Overview

This appears to be a transparent local safety-scanning tool; it reads selected files and can install a persistent Git pre-commit hook, but the artifacts do not show hidden exfiltration or destructive behavior.

This skill looks reasonable for local pre-publish safety checks. Before installing, be aware that the scanner reads the files you point it at, and the Git hook persists inside each repository and can replace/block the normal commit workflow.

Findings (3)

Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.

What this means

If run on a broad path, the skill will locally inspect many files and report filenames that may contain secrets, PII, or internal paths.

Why it was flagged

The scanner reads every matching file under the selected target path to look for secrets and PII. This is the core purpose, but it can inspect a large amount of local content if pointed at a broad directory.

Skill content
TARGET="${1:-.}" ... FILES=$(find "$TARGET" -type f ...); CONTENT=$(cat "$f" 2>/dev/null || continue)
Recommendation

Run the scanner only on files or repositories you intend to publish, and review the output before sharing it.

NoteHigh Confidence
ASI10: Rogue Agents
What this means

After installation, commits in that repository may be blocked until scan findings are fixed, bypassed with --no-verify, or the hook is removed.

Why it was flagged

The installer writes an executable Git pre-commit hook into the chosen repository, so the scanner continues running automatically on future commits.

Skill content
HOOK="$HOOK_DIR/pre-commit" ... cat > "$HOOK" << 'HOOKEOF' ... chmod +x "$HOOK"
Recommendation

Install the hook only in repositories where you want this guardrail, and back up any existing pre-commit hook before installing.

What this means

Running the health check touches private workspace or memory files and may include file-size or count information in the output.

Why it was flagged

The health check reads OpenClaw workspace and memory-related files to calculate sizes/token estimates. The artifacts show local counting, not printing file contents or transmitting them.

Skill content
WORKSPACE="${HOME}/.openclaw/workspace" ... GROUNDING_BYTES=$(cat "$WORKSPACE/THE_FRAMEWORK.md" ... "$WORKSPACE/STREAM.md" 2>/dev/null | wc -c)
Recommendation

Run the health check locally and review its output before sharing logs or screenshots.