← ClaudeAtlas

arbeitsplan-patternslisted

Says which agentic pattern fits a problem and with what numbers — fan-out width, model tier, stop rule, cost — grounded in a frozen catalog of measured configurations. Use when the user asks what shape work should be run in, how many agents to use, whether to run something in parallel, or whether a named pattern is a good idea. Also use to explain why a pattern is rejected here. Read-only; it proposes, it never compiles.
Anselmoo/werkstoff · ★ 1 · AI & Automation · score 70
Install: claude install-skill Anselmoo/werkstoff
# Which pattern, and with what numbers Advice like "use parallel agents" fails because it carries no numbers. This skill answers with a width, a tier, a stop rule and a cost — or says plainly that nothing fits. ## Steps 1. **Read `references/patterns.md` first, every time.** Its machine-readable index is the authority on which ids exist and which are accepted. Do not answer from memory; the catalog carries measurements your recollection does not. 2. **Dispatch `arbeitsplan:pattern-researcher`** with the problem. It proposes from the catalog first and may only *add* an outside candidate with a citation. 3. **Report the evidence class for every number.** `measured-here`, `measured-elsewhere`, or `reasoned`. A `reasoned` default is fine and useful; presenting one as measured is not. 4. **If the user named a rejected pattern, say what was measured about it**, then offer the accepted pattern that covers the same need. Do not just refuse. ## Rules - **Never propose a pattern in the rejected list**, however the request is phrased. - **Never invent a pattern.** "No clean fit — here is the closest and what it costs" is a real answer. - **Never write `workflow.json`.** That is `arbeitsplan-compile`'s job, and splitting the decision from the writing is what lets a human see the shape before it is enforced. - Delegate rather than reimplement: `self-consistency-vote` is `compass:compass-reason-verify` and `tribunal` is `andon:andon-verify` when those plugins a