tk-skill-diagnose
Pass
Audited by Gen Agent Trust Hub on Sep 3, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill operates entirely as a set of logical instructions for the AI agent to follow during debugging. It enforces a 'freeze' and 'evidence' workflow that prevents the agent from making arbitrary modifications to the source code (canonical skill) or jumping to conclusions without reproduction.
- [COMMAND_EXECUTION]: The skill mentions running experiments and checking git status (e.g.,
git check-ignore,git ls-files). These are standard developer tools used within a local checkout environment for diagnostic purposes and do not involve suspicious remote execution or privilege escalation. - [DATA_EXPOSURE_AND_EXFILTRATION]: While the skill analyzes incidents and traces, it explicitly instructs the user/agent: 'Do not copy raw logs, transcripts, screenshots, secrets, or repeated run narration.' This safety constraint minimizes the risk of sensitive data exposure during the reporting phase.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted 'incident prompts' as data. It mitigates this risk by requiring 'fresh execution in a clean context' and separating the diagnosis from the actual writing tool (
tk-learn), ensuring that any malicious instructions in the incident data do not automatically translate into system changes.
Audit Metadata