review-pull-request-qa
Pass
Audited by Gen Agent Trust Hub on Nov 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Pull Request code, project instructions, and external logs, which could contain malicious instructions designed to influence agent behavior.
- Ingestion points: Pull Request head/base revisions, project requirements, browser interaction outcomes, and server/worker logs (found in
SKILL.mdandreferences/vstack/capabilities/code-tests/REFERENCE.md). - Boundary markers: Employs a structured 'Quality Validation' protocol and a strict 'Definition of Done' to verify evidence and outcomes (found in
references/vstack/protocols/quality-validation.md). - Capability inventory: Executes shell-based test runners, database seed commands, and performs file system reads for logs and diagnostics.
- Sanitization: Explicitly mandates the redaction of sensitive data from logs and traces before the report is returned to the user.
- [COMMAND_EXECUTION]: The workflow requires executing database seed commands and browser automation tools which can be exploited if the local environment is misconfigured.
- Evidence:
SKILL.mdandreferences/vstack/capabilities/nextjs-tests/REFERENCE.mddescribe discovering and running seed commands, fixtures, and browser automation suites like Playwright or Cypress. - Mitigation: Includes pre-execution checks to verify that seeds are non-production, isolated, and reversible.
- [DATA_EXFILTRATION]: The skill accesses server logs and diagnostic traces which are potential sources of sensitive information leakage.
- Evidence:
SKILL.mdstep 4 describes correlating server/worker logs and browser console requests. - Mitigation: Contains specific instructions to redact sensitive data from these sources to prevent exposure in the final report.
Audit Metadata