bug-fix
Pass
Audited by Gen Agent Trust Hub on May 26, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it is designed to ingest and process untrusted external data that could contain malicious instructions.
- Ingestion points: SKILL.md Step 1 gathers observed behavior from failing tests, error messages, stack traces, logs, and request payloads.
- Boundary markers: The skill does not explicitly define markers to separate untrusted data from system instructions, although it references a
framework:secure-codingskill. - Capability inventory: SKILL.md Steps 2, 3, 5, and 6 involve executing automated tests, creating reproduction paths, and implementing code changes.
- Sanitization: No explicit sanitization or filtering of the ingested data is described in the workflow.
- [COMMAND_EXECUTION]: The skill requires the agent to execute code in the form of automated tests and reproduction scripts (Steps 2 and 3). While this is the intended primary purpose of the skill, the execution happens on the local environment and relies on the user's project context.
Audit Metadata