start-sessionlisted

Initializes the working session — reads the project's source of truth, the .crew memory (techstack, operations, hardenings), the real git state and the loose ends from the last session, in READ-ONLY mode, creates the memory files if missing, expires old plans, and confirms scope alignment and active gates before any code. Use at the start of every session.
mateusands/claude-code-crew-kit · ★ 0 · AI & Automation · score 62

Install: claude install-skill mateusands/claude-code-crew-kit

# Session start - **Can:** read the source of truth, the `{{RECORDS_DIR}}` memory and the real git state; create the memory files if they are missing; expire plans past their 7 days. - **Must:** read `info.md` fresh rather than from memory, state the mode it declares, and confirm scope and the active gates before any code. - **Cannot:** touch product code — the session opens READ-ONLY — and cannot delete anything under `hardenings/`. The goal right now is **not to code**: it is to load context and confirm we are in the right scope. Wrong context at the start costs the whole session. ## 1. Absorb the source of truth Read, in this order: 1. 🔴 **`{{RECORDS_DIR}}/info.md`** (authority) **and `.crew-kit-config`** (this machine's roster): who reviews, who decides each gate. **Read it fresh, never from memory of a previous session**, and state the mode in your opening report. In `solo` mode there is no cross-review, and every later report must say so. Then run the gate that checks the roster against the machine: ```bash node .claude/scripts/check-local-llm.mjs # exits 1 on an undeclared caller ``` It answers one question: is anything driving a LOCAL model automatically that the roster did not ask for. That state is silent by construction — it costs GPU and holds VRAM with nobody watching, and it was found once by noticing the machine was warm, which is not a method. A model loaded by hand is not a finding; an automatic caller is. The s