finish-feature

Pass

Audited by Gen Agent Trust Hub on Sep 12, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: Executes shell commands including git for version control operations and gh for GitHub CLI interactions. It also runs arbitrary scripts defined in the project's package.json file for linting, formatting, and testing phases.
  • [EXTERNAL_DOWNLOADS]: Downloads and executes linting and formatting utilities (oxlint, oxfmt) from the NPM registry via npx or bunx if the repository does not have its own configured tools.
  • [INDIRECT_PROMPT_INJECTION]:
  • Ingestion points: Reads project configuration files (package.json), documentation (CLAUDE.md), and repository diffs (gh pr diff).
  • Boundary markers: The skill interpolates the raw diff output directly into the prompt provided to the subagent review loop.
  • Capability inventory: The skill possesses capabilities for file system modification, command execution, and network communication via version control tools.
  • Sanitization: No specific sanitization or filtering is performed on the diff content or repository files before they are processed by the main agent or the review subagent.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 12, 2026, 03:19 AM
Security Audit — agent-trust-hub — finish-feature