project-stage-detect

Pass

Audited by Gen Agent Trust Hub on Aug 24, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data by scanning and reading project files (design documents, source code, production artifacts). This creates a surface for indirect prompt injection if those files contain malicious instructions aimed at the agent.
  • Ingestion points: Scans files in design/, src/, production/, prototypes/, docs/architecture/, and tests/ directories using Glob, Grep, and Read tools.
  • Boundary markers: The instructions do not define explicit delimiters (like XML tags) to separate project data from agent instructions during processing.
  • Capability inventory: The agent has access to Bash (local command execution) and Write (file creation) tools.
  • Sanitization: The skill relies on a 'Collaborative Protocol' and 'Request Approval' step before committing changes, which acts as a manual checkpoint but does not programmatically sanitize input data.
Audit Metadata
Risk Level
SAFE
Analyzed
Aug 24, 2026, 04:18 PM
Security Audit — agent-trust-hub — project-stage-detect