← ClaudeAtlas

starting-worklisted

The pre-work decision gate: inspects git state, recommends location, branch (feat/fix <slug>), session name (ADR-0087/0088), PR requirement, push target, and PR sequence (ADR-0072) for multi-unit scope — confirmed before any write. Invoke for "implement", "build", "fix", "refactor", even unnamed. Skip for research/questions.
monte3l/m3l-automation · ★ 0 · DevOps & Infrastructure · score 56
Install: claude install-skill monte3l/m3l-automation
# starting-work This skill is the single place the repo answers "where do I do this work?" before touching anything. That question used to be re-derived independently in four spots (`guard-branch-isolation.mjs`, `implementing-submodules` Step 0, `auditing`, `creating-prs`), and the scaffolding skills wrote guarded paths with no isolation step at all — so they hit `guard-branch-isolation.mjs` mid-run instead of branching proactively. Centralizing the decision here keeps the answer consistent and gets it made _before_ the first edit, which is the only time it's cheap to change. ## Why a gate at all `guard-branch-isolation.mjs` hard-blocks writes to `packages/*/src/**`, `scripts/*/src/**`, and `**/tests/**` while `HEAD` is `main`. That's a backstop, not a plan: if you discover it when a write is rejected, you're already mid-task with a dirty tree. Building on `main` left the working tree dirty for a whole run once (`docs/logs/2026-07-01-core-analysis.md`, divergence 7). This skill is the workflow half — it branches _before_ the block can fire — and it makes the branch/PR/push choices explicit so nothing silently lands on `main`. ## The contract **Infer and recommend all decisions, then confirm every one with the user in a single round. Do not write files, create a branch, or create a worktree until the user has confirmed.** The user is always free to override a recommendation; your job is to make the right default obvious, not to force it. Location, branch, session name, PR