technical-pm
Pass
Audited by Gen Agent Trust Hub on Mar 1, 2026
Risk Level: SAFECOMMAND_EXECUTIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface as it coordinates sub-agents that process untrusted external data and internal repository files. * Ingestion points: Reads WORK-LOG.md, .archive-metadata.yaml, and research data from sub-agents. * Boundary markers: Implements Integrity Sentinels and Handoff validation protocols for context passing. * Capability inventory: Access to Bash execution (via Task tool), file system operations, and WebSearch (via sub-agents). * Sanitization: Includes quality gates and mandatory user confirmation checkpoints for parallel execution and critical decisions.
- [COMMAND_EXECUTION]: The skill provides explicit instructions for the orchestrator to invoke sub-agents with the bash subagent_type via the Task tool for technical coordination tasks.
- [DYNAMIC_EXECUTION]: Employs a parallel execution protocol that dynamically generates instructions for sub-agents by substituting user-specific variables into predefined markdown templates at runtime.
- [CREDENTIALS_UNSAFE]: The assets/crisis-response-template.md file contains a hardcoded pattern matching sk-proj- within a documentation example meant for incident response training.
- [DATA_EXPOSURE]: The skill attempts to access the file path ~/.claude/skills/archive-workflow/references/archival-compliance-check.md to perform archival compliance verification.
Audit Metadata