stabilization-loop

Pass

Audited by Gen Agent Trust Hub on Mar 5, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill automates the execution of local development tools including compilers, linters, servers, and test suites.
  • [COMMAND_EXECUTION]: The agent is instructed to modify the local environment by committing changes and tagging the repository via Git.
  • [PROMPT_INJECTION]: The skill has an indirect prompt injection surface (Category 8) because the agent is instructed to analyze and act upon potentially untrusted data from test logs and server crashes.
  • Ingestion points: System output, test error logs, and server crash reports (SKILL.md).
  • Boundary markers: No explicit delimiters or instructions to ignore instructions embedded in logs are present.
  • Capability inventory: The agent can modify files, compile code, execute servers/tests, and perform Git operations (commit/tag).
  • Sanitization: No sanitization or validation of the log content is performed before the agent processes it for 'fixing' code.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 5, 2026, 10:54 AM