← ClaudeAtlas

worktree-proof-stacklisted

Move bounded coding-agent work from an initial idea to verified terminal closure with target checks, non-overlapping scopes, safe execution, and redacted evidence. Use when planning, delegating, running, integrating, or closing work that must remain auditable.
Nedal7707/worktree-proof · ★ 0 · Code & Development · score 70
Install: claude install-skill Nedal7707/worktree-proof
# WorktreeProof Stack Use this stack to move from a useful first idea to a verified, terminally closed result. Follow the vibe-to-verified path: keep the objective bounded, make ownership explicit, and treat every claim as pending until its evidence is present. ## Orient and bound 1. Read the applicable project instructions and identify the one objective to advance. 2. Verify the current working directory, repository identity, branch, and target revision before the first write. 3. State the relative file or resource boundary, the integration target, and the terminal outcome required for closure. 4. Set finite time, tool, and retry budgets. Record a blocker instead of waiting for an unbounded reply or extending a no-progress loop. ## Prove independence 1. Split work only into genuinely independent units. 2. Give each unit a unique normalized lane identifier and one non-empty, relative scope. 3. Reject duplicate, parent-child, traversal, and normalized-equivalent scopes before any lane starts. 4. Keep shared files, mutable records, external resources, and hidden state out of concurrent lanes. Give each build lane a named integration action. 5. Keep one accountable coordinator for decisions. Executors return artifacts and evidence; they do not approve, block, or supervise one another. ## Reserve and execute safely 1. Reserve the lane and scope atomically. If reservation fails, report the precise conflict and re-scope rather than forcing it. 2. R