control
Pass
Audited by Gen Agent Trust Hub on Aug 31, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests status and trust metrics from local files to determine the automation level of the agent. If these files are manipulated, it could influence the agent's autonomy.
- Ingestion points:
.bkit/runtime/control-state.jsonand.bkit/state/trust-score.json(SKILL.md). - Boundary markers: None present; the skill assumes the integrity of the local state directory.
- Capability inventory: The skill has access to
Bash,Write,Read,Glob, andGreptools. - Sanitization: The skill performs input validation (ensuring level is 0-4) and requires explicit user confirmation via
AskUserQuestionwhen escalating the automation level. - [COMMAND_EXECUTION]: The skill utilizes the
Bashtool and interacts with local JavaScript modules (lib/audit/audit-logger.js,lib/control/automation-controller.js) to implement its logic. This is standard behavior for a system management skill but involves executing local script logic.
Audit Metadata