hillclimbing
Pass
Audited by Gen Agent Trust Hub on Nov 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill dynamically generates and runs shell commands ('checks') to measure progress toward a goal. This is a core part of its intended functionality for automated task completion.
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted input from user-defined criteria and the output of executed commands, creating a potential surface for indirect prompt injection.
- Ingestion points: Goal/criteria definitions from the user; output from the execution of the 'check' command.
- Boundary markers: Absent; the instructions do not specify delimiters for external data.
- Capability inventory: Ability to execute arbitrary shell commands (the check) and write progress logs to the local filesystem.
- Sanitization: No explicit sanitization or validation of the criteria or command output is mentioned.
- [PERSISTENCE]: The skill records its progress and iteration logs in a file located 'outside the working tree'. This is used as a legitimate mechanism to maintain state across agent context resets and does not appear to be an attempt at malicious persistence.
Audit Metadata