bugfix
Pass
Audited by Gen Agent Trust Hub on Jul 25, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute various shell commands, test suites, and utility tools like curl or git bisect to create a deterministic reproduction of reported bugs. Evidence: Step 1 explicitly directs the agent to use commands, failing test invocations, or curl requests as a baseline for judging the fix.\n- [PROMPT_INJECTION]: The skill ingests untrusted data from bug reports and local logs, which creates a surface for indirect prompt injection. Ingestion points: User-provided symptoms via the /bugfix command and entries within the out/dev/bugfix-log.md file. Boundary markers: No specific delimiters or safety instructions are defined to isolate processed data. Capability inventory: The agent is granted shell command execution (Step 1 and Step 3) and file system write access for fixing code and logging (Step 4). Sanitization: The instructions do not prescribe sanitization or validation of the ingested symptoms or log data.
Audit Metadata