maintain-prove-it
Warn
Audited by Gen Agent Trust Hub on Nov 1, 2026
Risk Level: MEDIUMINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes potentially untrusted data from past execution receipts and git logs to modify its own instructions.\n
- Ingestion points: Verification receipts in
.prove-it/runs/,git logoutput, and user-provided bug descriptions.\n - Boundary markers: No specific delimiters or warnings are used to separate untrusted data from the agent's instructions.\n
- Capability inventory: The skill can modify its own
SKILL.mdandscripts/folder and then execute those changes.\n - Sanitization: No sanitization or verification of extracted commands is performed before they are written and executed.\n- [DYNAMIC_EXECUTION]: The skill performs runtime self-modification and execution of its own code.\n
- Evidence: The skill is explicitly instructed to "write in the working commands" derived from logs and then "prove every edit live" by running the modified start, health, and drive sequences.
Audit Metadata