review-work

Pass

Audited by Gen Agent Trust Hub on Jun 23, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATION
Full Analysis
  • [PROMPT_INJECTION]: The skill contains an indirect prompt injection surface as it processes untrusted content from source files, diffs, and external communication channels and passes them to sub-agents.\n
  • Ingestion points: File contents, git diffs, and retrieved content from Slack, Notion, and GitHub issues are used as prompt context.\n
  • Boundary markers: The skill employs XML tags (e.g., <file_contents>, , <original_goal>) to delimit external data from core instructions.\n
  • Capability inventory: The skill can spawn sub-agents, execute git/gh commands, and run user-defined application start commands.\n
  • Sanitization: It includes explicit instructions to redact sensitive information such as tokens, API keys, and PII before sharing review results.\n- [COMMAND_EXECUTION]: The skill uses shell commands for review orchestration and verification.\n
  • It uses git commands for context gathering and creates isolated git worktrees to avoid polluting the main environment.\n
  • The QA sub-agent is designed to execute the application's own build or run scripts to verify functionality.\n- [DATA_EXFILTRATION]: The skill reads data from local source code and well-known communication platforms.\n
  • It accesses GitHub, Slack, and Notion to ensure implementations align with historical decisions and documented requirements. This access is performed within the local agent context using standard integration tools.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 23, 2026, 11:37 AM
Security Audit — agent-trust-hub — review-work