ai-counterpart-discipline

Pass

Audited by Gen Agent Trust Hub on Sep 1, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill implements a robust 'Authority Charter' that explicitly lists reserved actions that only a human can perform (e.g., irreversible actions, budget extensions, and protected-branch merges). This prevents the AI from exceeding its intended boundaries.
  • [EXTERNAL_DOWNLOADS]: The SKILL.md file provides a command to install a missing dependency: npx skills add FuDesign2008/open-skills -g --skill intake-interview-discipline --yes. This refers to the author's own repository and a standard package management workflow, which is consistent with the skill's purpose and the author's identity.
  • [REMOTE_CODE_EXECUTION]: While the skill mentions the execution of shell commands for fetching facts ([FACT:cmd]), this is framed as a methodology for the AI agent to verify evidence during a review process within its existing environment, rather than a vector for downloading and executing untrusted external scripts.
  • [PROMPT_INJECTION]: The skill uses strong instructional language ('Mandate', 'Forbidden', 'Must NOT') to define safety boundaries. It does not attempt to bypass LLM safety filters; instead, it establishes internal logic for an 'anti-sycophancy' protocol to improve the quality of AI reviews.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 1, 2026, 03:39 AM
Security Audit — agent-trust-hub — ai-counterpart-discipline