full-review-pr

Pass

Audited by Gen Agent Trust Hub on Nov 7, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONMETADATA_POISONING
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingests and processes untrusted data from GitHub PRs and Jira tasks, creating an attack surface where malicious code or descriptions could attempt to influence the agent's behavior.
  • Ingestion points: SKILL.md (Workflow Step 1 and 3) describes reading the PR title, description, diff, and associated Jira issue content.
  • Boundary markers: The instructions for the Moderator and Analyst agents do not define clear delimiters (e.g., XML tags or specific markers) or instructions to ignore embedded commands within the analyzed code.
  • Capability inventory: The system can perform local checkouts via gh pr checkout, read file system content, and delegate tasks to other AI models.
  • Sanitization: The workflow does not specify any sanitization or escaping of the external content before passing it to the sub-agents.
  • [COMMAND_EXECUTION]: The skill uses the GitHub CLI to perform local operations.
  • Evidence: SKILL.md specifies the use of gh pr checkout <number> to bring PR code into the local environment for analysis.
  • [METADATA_POISONING]: The skill references non-existent or fictional model versions which may lead to unpredictable behavior depending on how the platform resolves these identifiers.
  • Evidence: references/moderator.md and references/pr-comments.md reference models such as gpt-6-astra, gpt-6.1-sol, claude-opus-5-5, and claude-fable-5-1.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 7, 2026, 02:31 AM