plan-pipeline-validate

Pass

Audited by Gen Agent Trust Hub on Aug 17, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted plan data and architectural decisions to drive validation logic and trigger domain-specific agents.\n
  • Ingestion points: Reads development plan files from docs/plans/ and architectural decision records (ADRs) from docs/adr/.\n
  • Boundary markers: The skill does not employ explicit delimiters or instructions to prevent the agent from following directives embedded within the plan documents.\n
  • Capability inventory: Spawns parallel specialist agents via a Task tool and programmatically modifies project documentation and plans.\n
  • Sanitization: No sanitization or intent validation of the ingested plan content is specified before it triggers agent behavior.\n- [COMMAND_EXECUTION]: Utilizes a background task management system to spawn and monitor multiple concurrent agent processes (e.g., security-reviewer, performance-reviewer).\n- [PROMPT_INJECTION]: Includes instructions to automatically transition to the /plan-pipeline:execute command without human confirmation if all issues match auto-resolve patterns, which could be exploited to bypass manual review of generated changes.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 17, 2026, 07:36 AM
Security Audit — agent-trust-hub — plan-pipeline-validate