cook-parallel

Pass

Audited by Gen Agent Trust Hub on Apr 29, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to execute a local utility script (python .claude/scripts/code_graph) for dependency mapping. This relies on the existence of a specific directory structure (.claude/scripts/) within the user's workspace.
  • [PROMPT_INJECTION]: The skill ingests user-supplied task descriptions via the $ARGUMENTS placeholder and processes external project documentation. To mitigate risks, the skill implements 'Rationalization Prevention' and 'Red Flag Stop Conditions' that force the agent to cite evidence and stop for security-sensitive code.
  • [SAFE]: The workflow explicitly mandates human-in-the-loop checkpoints (using AskUserQuestion) before dispatching sub-agents or moving past critical decision points. It includes clear instructions to stop if changes affect a large number of files or cross service boundaries.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 29, 2026, 09:18 AM
Security Audit — agent-trust-hub — cook-parallel