principle-prove-it-works
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill directs the agent to verify task completion by interacting directly with real artifacts, running features, and inspecting diffs. This creates a vulnerability surface where untrusted data or code being verified could contain instructions that influence the agent's behavior during the verification process.
- Ingestion points: The agent is instructed to read and verify external artifacts, source code, and diffs (described in
SKILL.md). - Boundary markers: The instructions lack specific guidance or delimiters to ensure the agent ignores potentially malicious instructions embedded within the artifacts it is inspecting.
- Capability inventory: The skill explicitly commands the agent to execute features, script checks, and utilize "Computer Use" tools to perform these verifications (referenced in
SKILL.mdandUPSTREAM.md). - Sanitization: No sanitization, filtering, or validation steps are prescribed for the content ingested during the verification process.
- [DYNAMIC_EXECUTION]: The skill encourages the agent to generate and execute scripts at runtime to perform deterministic checks, which involves the creation of executable content to verify the state of the system.
- Evidence:
SKILL.mdcontains the explicit instruction: "Script the check when you can. A deterministic script that reruns the comparison beats a one-time look."
Audit Metadata