skills/benjaming/ai-skills/pr-respond/Gen Agent Trust Hub

pr-respond

Pass

Audited by Gen Agent Trust Hub on Jun 29, 2026

Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses the GitHub CLI (gh api) to perform several actions on the GitHub platform, including posting comments, adding reactions, and resolving review threads. These operations are gated by a mandatory batch preview and a single-confirmation requirement from the user.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes data derived from GitHub Pull Request comments, which are external and untrusted. There is a theoretical risk that malicious instructions in a comment could influence the prior triage step, leading this skill to apply unintended 'fixes'. However, the design requires the agent to present a summary for user review before execution, providing a strong control against such attacks.\n
  • Ingestion points: Reads data (dispositions, thread IDs) from the conversation context produced by a previous triage step.\n
  • Boundary markers: Absent in the instructions.\n
  • Capability inventory: Modifies local source code and executes gh api network requests.\n
  • Sanitization: Relies on a 'humanizer' skill for drafting replies, which may provide some natural language filtering, but primarily relies on user confirmation of the planned batch actions.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 29, 2026, 11:53 PM
Security Audit — agent-trust-hub — pr-respond