← ClaudeAtlas

buildd-mcp-consumerlisted

Use whenever the buildd MCP tools (`buildd`, `recall`, `learn`) are available and you're about to act on a buildd task or file one — task coordination workflow: claiming, working, and completing tasks; deciding between a hard block and a flagged assumption; reporting friction; and which branch a task's PR should target.
buildd-ai/buildd · ★ 1 · AI & Automation · score 70
Install: claude install-skill buildd-ai/buildd
# Buildd MCP Consumer Skill You have the buildd MCP server mounted. This skill is the procedure for using it — the connector's `instructions` only tell you your token level and that this skill exists; everything about *how* to work a task lives here. ## Task Lifecycle ``` recall (check prior context) → claim_task → work → update_progress (milestones) → create_pr → create_artifact → learn → complete_task ``` 1. **Before starting:** `recall query="<task title>" scope=["memory","task"]` — surfaces prior gotchas, patterns, and how similar tasks were solved. 2. **Claim:** `buildd action=claim_task`. Response includes your worker ID (save it — every later call needs it, though most auto-resolve it from context), the branch to check out, and the task description. 3. **Work it.** Check out the returned branch and make the change. 4. **Report progress at milestones** (roughly 25%, 50%, 75%): `buildd action=update_progress params={ progress, message }`. 5. **Ship:** push commits, then `buildd action=create_pr`. Optionally `action=get_pr` to check CI/review state, then `action=merge_pr` once green — merging is subject to the workspace's merge policy tier, and a 403 there names the reason; don't retry blindly. 6. **Before completing:** `buildd action=create_artifact` with a summary of what you did (`type=summary`), or `type=impl_plan` with a `key` if a later task is meant to read this one's output. Then `learn` anything a future agent would want to kno