← ClaudeAtlas

buildlisted

Implements code incrementally with quality gates. Use when the user says 'build' or 'implement', or when starting the implementation phase of an approved plan.
domengabrovsek/agent-config · ★ 17 · AI & Automation · score 70
Install: claude install-skill domengabrovsek/agent-config
Implement the approved plan incrementally: $ARGUMENTS Follow these disciplines: ## Before Starting **why-no-hook:** skill workflow guidance; each step requires understanding the surrounding context (repo, task shape, prior state). - Verify an approved plan exists (in `.claude/state/plans/` or the current conversation) `(review-time: see section note)` - If no plan exists, run `/plan` first. With no spec either, run `/spec` `(review-time: see section note)` - Read the plan and identify the task list `(review-time: see section note)` - Copy the task list to `.claude/state/runs/<branch-slug>/tasks.md` and tick each task as it finishes. Read that file, not memory, to resume after compaction `(review-time: see section note)` - If a spec drives the plan, set its `branch:` frontmatter to the current branch `(review-time: see section note)` - Spawn the matching expert teammates for the plan's domain (see `rules/agent-routing.md`) - their guardrails apply to every increment `(review-time: see section note)` - If the plan has 2+ file-isolated lanes, execute it in **lane mode** - spawn one lane-mode teammate per lane (see `rules/parallel-agents.md`); single-lane plans stay in this session `(review-time: lane-vs-single judgment from the plan shape)` ## Increment Rules For each task in the plan: 1. **Ask**: "What is the simplest thing that could work?" `(review-time: see section note)` 2. **Scope**: touch only what the task requires - no drive-by refactors, no "while I'm here" chan