rpi-plan-critique

Pass

Audited by Gen Agent Trust Hub on Sep 8, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • Indirect Prompt Injection Surface: The skill is designed to ingest and critique implementation plans and evidence from external file paths provided by the caller. This creates a potential surface where instructions hidden within these files could attempt to influence the agent's behavior or assessment logic.
  • Ingestion Points: The skill reads data from files specified by the plan and evidence arguments (SKILL.md).
  • Boundary Markers: The instructions direct the agent to explicitly define input and criterion boundaries before starting the assessment, which serves to scope the analysis (SKILL.md, Flow step 4).
  • Capability Inventory: The skill has the ability to read workspace files and write a critique artifact to a caller-specified output path.
  • Sanitization: The instructions include strict constraints that prohibit the agent from performing open-ended research or inferring missing evidence as fact, which helps mitigate the influence of external content (SKILL.md, Flow step 3).
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 8, 2026, 10:54 AM
Security Audit — agent-trust-hub — rpi-plan-critique