resume
Pass
Audited by Gen Agent Trust Hub on Sep 12, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests data from external files and git history that could contain adversarial instructions designed to influence agent behavior during experiment resumption.
- Ingestion points: Reads configuration (config.cfg), strategy (program.md), and history (results.tsv) files, as well as git log output (SKILL.md, Step 2).
- Boundary markers: The skill does not use specific delimiters or instructions to ignore potential commands within the ingested data.
- Capability inventory: The skill executes local Python scripts, performs git operations, and can trigger downstream iteration commands like /ar:run and /ar:loop.
- Sanitization: No evidence of sanitization or validation of the ingested file content before it is processed by the agent.
- [COMMAND_EXECUTION]: The skill executes shell commands to interact with the repository and manage experiment state.
- Evidence: Uses git checkout, git log, and python to run skill-specific scripts (SKILL.md, Steps 1 and 2).
Audit Metadata