Skills
skills/edwinhu/workflows/dev-implement/Gen Agent Trust Hub

dev-implement

Pass

Audited by Gen Agent Trust Hub on Jun 22, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONREMOTE_CODE_EXECUTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses dynamic context injection (the ! syntax) to execute a local Python script (load-constraints.py) to initialize the agent context. It also provides instructions for using the Bash tool to perform testing and monitoring operations during the development lifecycle.
  • [PROMPT_INJECTION]: The skill processes untrusted data from multiple sources including .planning/PLAN.md, .planning/SPEC.md, and outputs from sub-agents, which creates a surface for indirect prompt injection. Ingestion points: .planning/PLAN.md, .planning/SPEC.md, .planning/PLAN_REVIEWED.md, LEARNINGS.md, VALIDATION.md, and output from Workflow/Agent tools. Boundary markers: None identified. Capability inventory: Bash, Agent, Workflow. Sanitization: None identified.
  • [REMOTE_CODE_EXECUTION]: The skill configuration includes PreToolUse hooks that execute local Python scripts (dev-delegation-guard.py and phase-gate-guard.py) using uv run to enforce security boundaries and workflow gates.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 22, 2026, 06:30 AM
Security Audit — agent-trust-hub — dev-implement