run-test-phase
Warn
Audited by Gen Agent Trust Hub on Apr 8, 2026
Risk Level: MEDIUMCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill constructs shell commands by interpolating feature names and file paths directly from
test-plan.mdintonpx vitest run. In Step 4.2, the execution ofnpx vitest run {file1} {file2} ...lacks input sanitization or robust quoting for variables derived from the externaltest-plan.mdfile. If the test plan contains shell metacharacters (e.g.,;,&,|), it could lead to arbitrary command execution on the host machine. - [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface. It reads
test-plan.md, source code, and test results, then passes this content to aqa-specialistsubagent without using boundary markers or sanitization. Malicious instructions embedded in the analyzed code or test plan could influence the subagent's behavior. - Ingestion points:
test-plan.md, test files, source files, andbackend-plan.md(Step 6). - Boundary markers: Absent. The prompt in Step 6 interpolates file content directly.
- Capability inventory: Shell command execution (
npx vitest,npx supabase), file system writes (test.result.json), and subagent invocation. - Sanitization: Absent. No escaping or validation of the ingested file content is performed before passing it to the subagent.
Audit Metadata