delegate

Pass

Audited by Gen Agent Trust Hub on Aug 14, 2026

Risk Level: SAFE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The README documentation provides installation instructions using npx and git clone targeting the author's own GitHub repository (github.com/kylezantos/fable-opus-delegate).- [INDIRECT_PROMPT_INJECTION]: The skill defines a workflow where an orchestrator model processes data (code diffs and tool outputs) generated by subagent models.
  • Ingestion points: The orchestrator reads diffs and execution logs reported by subagents in SKILL.md.
  • Boundary markers: The instructions explicitly command the agent to "verify — don't trust" and not to accept subagent claims without independent validation.
  • Capability inventory: The skill uses the Agent tool to delegate tasks and requires shell access to run verification commands (linting, testing).
  • Sanitization: The skill implements mitigation by requiring the orchestrator to review all changes and prohibits subagents from executing git commands that could hide or revert changes (git stash, git reset, etc.).
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 14, 2026, 08:51 PM
Security Audit — agent-trust-hub — delegate