← ClaudeAtlas

cooklisted

Keyword-driven orchestrator for this repository's coding standards. Receives a code-task summary from the invoking agent, extracts keywords, matches them against the global concern index and the relevant domain indexes, loads only the matched rules, and returns a single compiled standards payload.
ndisisnd/cook · ★ 1 · Data & Documents · score 65
Install: claude install-skill ndisisnd/cook
# Cook Orchestrator Cook is the single entry point for standards. It does not hold rules of its own — it detects what a task touches and composes a payload from the rule libraries under `standards/`. There are no mode flags (`--frontend` / `--backend` / `--full-stack`); layer concerns are matched by keyword, so a task that touches both UI and server simply matches concerns and domains from both sides. ## Protocol > **Phase 3 status (cook-feat-robust):** Steps 1–5 are the cache-first control > flow from Phase 1. Step 6 is the mechanical compilation script (Phase 2) — no > LLM assembly, deterministic, model-free on cache hits — now followed by a > mechanical self-heal of the `degraded` flag (Phase 3, feature 7). A corrupt > cache degrades to greedy routing (Component 8) without dropping standards. **Mode branch table.** Step 0 selects which path to take before anything else: | Mode | Condition | P0 loaded? | Cache participates? | Steps | |---|---|---|---|---| | `auto` | No args | **yes** (unconditional) | yes (signals fingerprint) | 1 → 2 → 3 → 4 → 5 → 6 | | `explicit-flags` | Flags only, no prose | **no** | yes (signals + flags fingerprint) | 0a → 0b → 0c → 5 → 6 | | `explicit-prose` | Any prose (± flags) | **no** | **no** (LLM must run anyway) | 0a → 0b → 0c → 6 | ### Step 0 — Parse user args Scan the invocation string **before** doing anything else: - Tokens beginning with `--` are **flags** of two forms: - **Simple flags** (`--security`, `--react`, `--global`): va