skill-creator
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill uses Python's
subprocessmodule across several scripts (run_eval.py,improve_description.py,generate_review.py) to invoke theclaudeCLI and system utilities likelsof. This execution is intended for evaluating skill triggering and managing a local evaluation viewer server. - [INDIRECT_PROMPT_INJECTION]: As a tool that ingests user-provided intent to generate and refine agent instructions, the skill has an inherent surface for indirect prompt injection. This is mitigated by the skill's design, which emphasizes iterative human-in-the-loop review and quantitative benchmarking.
- [EXTERNAL_DOWNLOADS]: The skill metadata and documentation reference official repositories from a trusted organization (Anthropic). These references are used for attribution and documentation and do not involve unverified remote code execution.
- [DATA_EXFILTRATION]: The
package_skill.pyscript implements proactive security measures to prevent data exposure. It uses an allowlist-based approach for packaging, specifically skipping common secret files (like.env,.pem, and SSH keys) and refusing to follow symlinks that resolve outside the skill directory.
Audit Metadata