parallel-test-fixing
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill workflow involves extracting failure output from testing frameworks and providing it to subagents as context for automated fixes.\n
- Ingestion points: Test failure output is captured from stdout/stderr during the execution of commands like
npm testorpytestas described inSKILL.md.\n - Boundary markers: The subagent task template in
SKILL.mddoes not utilize robust delimiters or instructions to prevent the subagent from interpreting content within the error logs as new commands.\n - Capability inventory: Subagents are authorized to read files, write to source code, and execute shell commands to verify their fixes in
SKILL.md.\n - Sanitization: The workflow described in
SKILL.mdlacks a sanitization or validation step to filter potentially malicious instructions embedded within the test output before it is passed to the subagent context.
Audit Metadata