incident-response-smart-fix
Pass
Audited by Gen Agent Trust Hub on Feb 27, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection as its primary function involves processing external data. Ingestion points: Phase 1 and Phase 2 prompts in resources/implementation-playbook.md ingest logs, observability data, and git commit history. Boundary markers: Absent; the instructions do not use delimiters or specify that analyzed content should not be interpreted as commands. Capability inventory: The skill can execute local scripts via 'git bisect run' and write code modifications to files. Sanitization: Absent; untrusted data is interpolated directly into prompts for subagents.
- [COMMAND_EXECUTION]: The workflow includes the 'git bisect run' command in Phase 2 to automate the discovery of regressive commits. This results in the automatic execution of local scripts, representing a powerful system-level capability that is part of the skill's core functionality.
Audit Metadata