execute-plan
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill manages the risk of malicious instructions embedded in GitHub issues or comments through several defensive layers. 1. Ingestion points: Reads issue bodies and plan comments from GitHub (SKILL.md, Section 1). 2. Boundary markers: Utilizes a mandatory '## Implementation Plan' structure and a 'Requirements SHA256' fingerprint to verify plan integrity. 3. Capability inventory: Performs git operations, file modifications, and executes local verification scripts (plan-header.mjs, delivery-status.mjs). 4. Sanitization: The plan-header.mjs script enforces strict schema validation and cryptographic consistency checks on input data before allowing the execution to proceed.
- [SAFE]: No obfuscation, data exfiltration, or unauthorized remote code execution patterns were detected. The included Node.js scripts are used for local data processing without network access.
Audit Metadata