better-skill-review

Pass

Audited by Gen Agent Trust Hub on Mar 6, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local Python and shell scripts (scripts/validate.py and scripts/analyze.sh) to perform automated validation and profile extraction of the target skill directory.
  • [PROMPT_INJECTION]: This skill presents an indirect prompt injection surface as it is designed to ingest and review content from external, potentially untrusted skills.
  • Ingestion points: In Step 5 (Semantic Review), the agent reads the full content of the target skill's SKILL.md and associated files.
  • Boundary markers: The automated analysis provides structured JSON output, but the semantic analysis relies on reading raw markdown without specialized delimiters.
  • Capability inventory: The agent can execute shell commands, Python scripts, and has file-writing capabilities through the 'Interactive Improvement' step (Step 7).
  • Sanitization: While the linter checks for secrets, it does not sanitize the target skill's instructions before they are processed by the LLM.
  • [SAFE]: The linter script (validate.py) includes specific logic to detect security risks in the skill under review, such as hardcoded API keys (OpenAI, GitHub, AWS), Slack tokens, and PII (emails), which is a security best practice for development tools.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 6, 2026, 03:53 AM