review-fix-loop

Pass

Audited by Gen Agent Trust Hub on Sep 3, 2026

Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses shell commands to manage agent threads and resolve project environments.
  • Evidence: Usage of bb-cli and bb thread wait <id> --timeout 1200 --json for environment resolution and worker monitoring.
  • Context: These commands are used to interact with the platform's native tooling for managing agent lifecycles and Git worktrees.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external data from issue trackers or specification documents, which could serve as a vector for indirect injection.
  • Ingestion points: Worker prompts in SKILL.md incorporate content from "attached spec or ticket" references.
  • Boundary markers: The prompts lack explicit isolation delimiters (e.g., XML tags or specific headers) between instructions and the untrusted external content.
  • Capability inventory: The skill can execute shell commands, spawn subagents, and commit code changes to the repository.
  • Sanitization: The skill implements a verification step where a separate "finding checker" must confirm citations against the source code before a fix is attempted, providing a logical gate against spurious instructions.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 3, 2026, 05:50 PM
Security Audit — agent-trust-hub — review-fix-loop