fix-pr-review

Pass

Audited by Gen Agent Trust Hub on Apr 23, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It retrieves PR review comments from external sources like GitHub APIs or local files in Phase 2 and passes them to a subagent in Phase 3. Ingestion points: GitHub GraphQL API, REST API, and local markdown files. Boundary markers: The subagent prompt lacks robust delimiters around untrusted comment bodies. Capability inventory: The agent has access to Read, Grep, Bash, Edit, and Write tools which can modify the repository. Sanitization: No sanitization of comment content is performed before interpolation into subagent instructions.
  • [COMMAND_EXECUTION]: The skill performs various shell operations through the Bash tool and local scripts. Evidence: Uses the gh CLI for GitHub operations, git for source control tasks, and development tools like tsc and turbo for type-checking during the fix validation phase.
  • [DATA_EXFILTRATION]: The skill accesses repository-specific data to perform its functions. Evidence: It fetches PR metadata, code diffs, and review comments via GitHub APIs. While required for functionality, this involves processing external and potentially sensitive repository data.
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 23, 2026, 09:14 PM