diagnose
Pass
Audited by Gen Agent Trust Hub on Apr 29, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill instructions define a standard software debugging lifecycle (Reproduce, Minimise, Hypothesise, Instrument, Fix, Test). No malicious intent or bypass attempts were detected.\n- [COMMAND_EXECUTION]: The skill facilitates the execution of local development tools (e.g., unit tests, curl, git, Playwright). These are used strictly within the context of reproducing and fixing bugs as requested by the user.\n- [DATA_EXFILTRATION]: While the skill involves processing bug reports, log files, and captured traces, it does not include hardcoded credentials or automated exfiltration logic. The data handled is provided by the user for the purpose of diagnosis.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources.\n
- Ingestion points: Phase 1 (logs, traces, payloads), Phase 2 (repro output), and hitl-loop.template.sh (user responses).\n
- Boundary markers: None identified in instructions or scripts.\n
- Capability inventory: Subprocess execution for testing and debugging, file-system access for logging, and debugger/REPL usage (Phase 4, Phase 5).\n
- Sanitization: None identified; the agent is expected to interpret data as debug information. This is an inherent risk of debugging tools but is handled within safe bounds.
Audit Metadata