skill-design-loop

Pass

Audited by Gen Agent Trust Hub on Jul 30, 2026

Risk Level: SAFE
Full Analysis
  • [SAFE]: The skill implements strong defensive configurations, including 'disable-model-invocation: true' in its frontmatter and a requirement for explicit invocation, which prevents unauthorized or accidental activation.
  • [COMMAND_EXECUTION]: The skill includes a local validation script ('scripts/self-test.mjs') intended for developers to verify the orchestration logic. This script uses standard Node.js 'fs' modules to read internal skill files and fixtures without executing external or untrusted code.
  • [DATA_EXFILTRATION]: Repository resolution is performed locally using 'git rev-parse --show-toplevel'. All write operations are restricted to the repository's '.skill-proposals/' directory, and no network exfiltration patterns were identified.
  • [PROMPT_INJECTION]: The skill presents a low surface for indirect prompt injection by defining a strict YAML schema for intake and resolving dependencies solely from the local repository. Evidence chain: 1. Ingestion points: Free-form skill ideas and checklists in SKILL.md. 2. Boundary markers: Strict YAML checklist normalization and repo-local source-of-truth rules. 3. Capability inventory: Local file reads/writes and repository discovery in SKILL.md. 4. Sanitization: Logic and routing validation provided by scripts/self-test.mjs.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 30, 2026, 09:29 AM
Security Audit — agent-trust-hub — skill-design-loop