The Agent Skills Directory

[SAFE]: The skill performs expected repository maintenance tasks such as auditing skill files (SKILL.md), checking for drift against source files using git log, and providing linting for metadata fields. All actions described in the Procedure and Report format are consistent with its stated purpose of preventing 'skill rot'.
[DATA_EXFILTRATION]: While the skill accesses file paths like .agents/distill-skill-ledger.json and optionally reads a usage log from ~/.claude/skill-usage.jsonl, these are standard locations for AI agent metadata and telemetry. There is no evidence of these paths being transmitted to external servers; the operations are purely local for the purpose of classification and reporting.
[COMMAND_EXECUTION]: The skill uses git log to detect code drift. This is a standard, safe use of version control tools. The skill documentation explicitly states that it should not perform 'silent body rewrites' and that critical actions like PRUNE require explicit user confirmation, adhering to the principle of least surprise and user control.
[PROMPT_INJECTION]: No patterns associated with prompt injection, safety bypasses, or instruction overrides were detected in the skill instructions or metadata. The 'Hard principle' section emphasizes non-destructive behavior by default.
[EXTERNAL_DOWNLOADS]: The skill does not download external scripts or packages. It relies on existing repository files and local Git history.

skill-audit