clean-code
Warn
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONDYNAMIC_EXECUTIONPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill contains a detailed mapping of agent roles to specific shell commands that invoke Python scripts (e.g.,
python .agents/skills/api-patterns/scripts/api_validator.py .). This provides the agent with explicit instructions to execute code from the filesystem. - [DYNAMIC_EXECUTION]: The skill creates a runtime dependency on executable logic stored at relative paths (e.g.,
.agents/skills/vulnerability-scanner/scripts/security_scan.py) that are external to the skill's own content. The integrity and contents of these scripts cannot be verified within the scope of this skill definition. - [PROMPT_INJECTION]: The instructions utilize high-urgency labels ('CRITICAL', 'VIOLATION') and red circle emojis to strongly influence agent behavior, aiming to override default task completion logic and enforce a rigid script-running workflow that may bypass user intent or standard safety protocols.
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to ingest and parse the output of external scripts, creating a surface where malicious script output could manipulate the agent's summary or subsequent logic.
- Ingestion points: Output from validation scripts like
ux_audit.py,security_scan.py, andlighthouse_audit.py(SKILL.md). - Boundary markers: None present to delimit script output or warn the agent about potentially malicious content in the output.
- Capability inventory: Shell command execution via Python, and instructions to summarize results and ask for confirmation (SKILL.md).
- Sanitization: No instructions provided for sanitizing or validating the output generated by the external scripts before it is processed by the agent.
Audit Metadata