← ClaudeAtlas

forge-worktreelisted

This skill should be used to run Context Forge builds in parallel across terminals — phrases like "forge-worktree", "build units in parallel", "work on two specs at once", "open another build in a new terminal", "claim unit 7", or "clean up the worktree". One unit = one git worktree = one branch = one terminal: it checks the unit is dependency-ready, claims it atomically (visible across all worktrees), creates the worktree, and hands the user the exact commands for the new terminal. Also lists claims and releases finished ones.
yerros/context-forge · ★ 0 · AI & Automation · score 75
Install: claude install-skill yerros/context-forge
# forge-worktree Parallel builds without collisions. Two sessions in one working tree WILL crash into each other (shared test suite, shared git state, shared context files) — so parallelism means isolation: each unit gets its own worktree and branch, and a claim that every other terminal can see. ## Argument - `/forge-worktree unit 07` (or `7 payments-crud`) → claim + create the worktree for that unit. - `/forge-worktree list` → active claims and their ages. - `/forge-worktree done 07` → release the claim, remove the (clean) worktree. No argument → show `list` and offer to claim the next ready unit. ## Claiming a unit (`new`) 1. **Dependency gate.** Read `context/specs/00-build-plan.md`: the unit may only be claimed if every unit it depends on is in `## Completed`. Parallel work is for **independent** units — if it depends on something still in flight, refuse and say which unit blocks it. Its spec must exist (else: `forge-spec` first). 2. **Claim + create** (atomic, shared across worktrees): ```bash bash "${CLAUDE_PLUGIN_ROOT}/skills/forge-worktree/scripts/forge-worktree.sh" new <NN> <slug> ``` A double claim loses cleanly ("ALREADY CLAIMED" + who/where). The script prints the exact next step — relay it verbatim to the user: open a **new terminal** → `cd <worktree-path> && claude` → `/forge-build unit NN`. 3. Remind: each parallel session burns quota independently — this buys wall-clock time, not tokens. Two or three parallel units is t