What problem does it solve?
This Skill reduces end-to-end turnaround time for engineering changes by routing work to the most effective AI tool (Codex, Kimi, or Claude) while enforcing pre-task and post-task quality gates through iCPG and Mnemos.
Core Features & Use Cases
- Tool-aware delegation with complexity scoring: Chooses between Kimi solo, Kimi + Codex auto-review, or Claude direct handling based on bounded risk dimensions (depth, fan-out, security boundary crossings, concurrency, and invariants).
- Codex stop-hook auto-review: Runs an automatic Codex review after tests pass and feeds high-severity findings back to Claude.
- Mandatory iCPG + Mnemos context transfer: Enforces duplicate checks, constraints, and risk queries before changes, and records drift/status after changes while transferring state across delegated agents via Mnemos checkpoints.
- Use Case: When you ask to “review the auth middleware for timing attacks” or “push to git,” the system routes to the right model, injects the right invariants, runs TDD/test loops, and ensures recorded symbol/risk tracking across tool boundaries.
Quick Start
Tell Claude to perform your repo change (for example, “review and fix the timing attack risk in the authentication middleware”), and let it delegate to the appropriate agent(s) automatically.
