Mariomarquezt/multi-flow-skills / skills/multi-flow
multi-flow
UnverifiedOur own end-to-end build flow for a NEW project — interview, multi-model plan, spec, tickets, multi-model build, multi-model review, multi-model fix, wrap. It chains grill-with-docs, to-spec and to-tickets (Matt Pocock's skills, called as they are, never edited) with our four skills (multi-plan, multi-build, sliced-review, multi-fix) and runs `/wrap lite` after each of ours. Use when the user says "/multi-flow", "new project from start to finish", "run the whole flow", or starts a new app and wants every phase governed. NOT for one step alone (use that step's skill) and NOT for fixing a single bug.
Multi Flow
Build a project from an idea to reviewed, fixed, committed code in one governed run.
This is OUR skill. It is the multi-model sibling of Matt Pocock's full-build (we do not edit his).
Two rules that keep it ours:
- Matt's skills are called, never copied or edited —
grill-with-docs,to-spec,to-tickets(andsetup-matt-pocock-skills). An upstream update would erase any change. If we ever write our ownto-specorto-tickets, we swap the name in the phase table below and nowhere else. - Model versions live in ONE file:
~/.claude/skills/_shared/models.md. This skill never names a model version. Say "Sol", "Grok" — read the exact id from that file.
The one idea: state lives in .multiflow/STATE.md and git, never in the conversation. Every
phase ends by writing state. A fresh agent reads state and continues.
The phases (one todo per phase; never start one before the one before it is done)
| # | Phase | Skill it calls | Whose | After it |
|---|---|---|---|---|
| 0 | Preflight | setup-matt-pocock-skills (only if no tracker yet) | Matt | — |
| 1 | Interview | grill-with-docs | Matt | — |
| 2 | Plan (stack, layout, risks) | multi-plan (mode 3) | ours | /wrap lite |
| 3 | Spec | to-spec | Matt | — |
| 4 | Tickets | to-tickets | Matt | — |
| 5 | Build | multi-build (mode 3), one run per ticket or frontier batch | ours | /wrap lite |
| 6 | Review | sliced-review (mode 3, no fix) | ours | /wrap lite |
| 7 | Fix | multi-fix (mode 3), fed by the review's findings.json | ours | /wrap lite |
| 8 | Close | wrap (full) | ours | — |
Phase 0 — Preflight
- Must be a git repo. If not, stop and ask. Create a working branch. Never run on the default branch.
- No issue tracker + triage labels configured? Invoke
setup-matt-pocock-skillsfirst (Phases 3–4 need it). - Write
.multiflow/STATE.mdfrom the template inREFERENCE.md. Done when: STATE.md exists, names the branch and tracker, every phase row is[ ].
Phase 1 — Interview
Invoke grill-with-docs. Do not summarise the interview as the spec — Phase 3 does that.
Done when: every question is answered, ADRs and glossary are written; list their paths in STATE.
Phase 2 — Plan with the models
Invoke multi-plan (mode 3) with the interview's output as the idea. It drafts the layout, has
critics attack it, rules on every finding, and ends with the agreed layout + "open fights".
The user settles the open fights before moving on. Record the agreed layout path and the
decision-log path in STATE.
Then run /wrap lite.
Done when: the layout is agreed, open fights are settled, wrap lite ran.
Phase 3 — Spec
Invoke to-spec. Give it the interview AND the agreed layout; it must not re-interview.
Done when: the spec is at the path/issue in STATE, names the test seams, and the user confirmed the seams.
Phase 4 — Tickets
Invoke to-tickets against the spec. Output is a dependency graph with blocking edges. Never flatten it.
Copy the blocked by column into STATE verbatim — never re-derive it.
Done when: the user approved the breakdown and every ticket is in STATE with number, blockers, status [ ].
Phase 5 — Build (our multi-build)
Work the frontier: tickets whose blockers are all [x], lowest number first. Each multi-build
run takes one ticket (or a batch of frontier tickets that touch different files).
- The ticket IS the spec. Hand
multi-buildthe ticket text as its frozen spec. Its Step 1 then only checks the acceptance criteria are testable; it does not rewrite them. Its Steps 2–8 (seams, split by file, builders, cross-review, integrate, one commit) run as written. - One commit per ticket run, message names the ticket number. Commit BEFORE ticking STATE.
- After EACH
multi-buildrun:/wrap lite(it runs by default — do NOT passno wrap). Done when: every ticket row is[x]with a commit hash and the full suite passes once.
Phase 6 — Review (our sliced-review)
Invoke sliced-review mode 3 over the built code. Read its FIELD-NOTES before the first run;
do not restate its slicing or ledger rules. Keep findings.json and the triage output under
.multiflow/review/. Then /wrap lite.
Done when: the review finished, triage is done, findings.json is saved.
Phase 7 — Fix (our multi-fix)
- You read every confirmed finding yourself and mark it
fix,deferorrejectwith a one-line reason. A reject needs evidence, not an opinion. - Invoke
multi-fixmode3on thefixset. HIGH first as its own run; MEDIUM and LOW may share one. - Re-run the full suite and typecheck. Read the diff like a contributor PR.
Then
/wrap lite. Done when: the suite is green and every HIGH is fixed with a proving test or deferred in STATE with a reason.
Phase 8 — Close
Invoke wrap (full mode). Done when: wrap finished and STATE shows every phase [x].
Rules for chaining our four skills
/wrap literuns at the end of each of our four skills by default. In this flow, leave it ON between them. Passno wrapONLY on the run right before Phase 8, because Phase 8 does the full wrap.- Each of our skills reads models from
_shared/models.md. Do not override a model here. reportmode is off unless the user asks.
Resuming after a reset
Read .multiflow/STATE.md and the last five commits. Pick the first unticked phase — inside Phase 5,
the lowest-numbered ticket whose blockers are all [x]. Do not re-read the conversation. Do not redo ticked work.
Red flags — stop
- About to edit a Matt skill (
to-spec,to-tickets,grill-with-docs) → don't. Change THIS file's phase table instead. - About to write a model version into this file → put it in
_shared/models.md. - About to start a ticket whose blockers are not all
[x]→ it is not on the frontier. - About to tick a ticket before its commit exists → commit first.
- About to skip
/wrap litebetween our skills → run it. - About to start Phase 7 without reading every finding yourself → read them.
- About to run Codex through the
codex:codex-rescuesubagent → banned. Usemulti-build's own Codex route.
Keep the prompt cache warm
Same rule as our other skills: no subagent tool call waits longer than 230 seconds. See
~/.claude/skills/sliced-review/references/keep-warm.md and put the rule in every subagent brief that waits.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| rea-changelog-updatemorluto/reaPrepare or rewrite REA release changelogs and GitHub release notes from pinned Git history, with verified contributor thanks and Release Please synchronization. | Stars: 469k | Updated | Claude CodeCodex | Verified | Install |
| academy-guideanthropics/skillsStop and check this skill before finishing any reply to a question about how to use Claude or a Claude product — it recommends matching courses, tutorials, and use cases from Claude Academy (academy.claude.com), Anthropic's learning hub. Trigger on: "how do I", "how can I", "getting started with", "what can Claude do", "teach me", "learn to use"; questions about artifacts, projects, skills, plugins, connectors, MCP; requests about rolling Claude out to a team, class, or organization; and any ask for training materials, onboarding content, or learning resources. Use it when the user is learning how to use a feature or product — not when they are mid-task and just want the task done. This skill composes with other skills: after consulting product documentation to answer how a Claude feature works, also check here for a matching course or tutorial — a docs-grounded answer and an Academy recommendation belong together. Only recommend on a strong match; never invent Academy content. | Stars: 197k | Updated | Claude CodeCodex | Verified | Install |
| claude-apianthropics/skillsReference for the Claude API / Anthropic SDK — model ids, pricing, params, streaming, tool use, MCP, agents, caching, token counting, model migration. TRIGGER — read BEFORE opening the target file; don't skip because it "looks like a one-liner" — whenever: the prompt names Claude/Anthropic in any form (Claude, Anthropic, Fable, Opus, Sonnet, Haiku, `anthropic`, `@anthropic-ai`, `claude-*`, `us.anthropic.*`, `[1m]`); the user asks about an LLM (pricing/model choice/limits/caching) — never answer from memory; OR the task is LLM-shaped with provider unstated (agent/MCP/tool-definition/multi-agent/RAG/LLM-judge/computer-use; generate/summarize/extract/classify/rewrite/converse over NL; debugging refusals/cutoffs/streaming/tool-calls/tokens). SKIP only when another provider is being worked on (overrides all triggers): OpenAI/GPT/Gemini/Llama/Mistral/Cohere/Ollama named in the query; OR `grep -rE 'openai|langchain_openai|google.generativeai|genai|mistralai|cohere|ollama'` over the project hits (run this grep FIRST if no provider named — don't Read the file). | Stars: 197k | Updated | Claude CodeCodex | Unverified | Install |
| doc-coauthoringanthropics/skillsGuide users through a structured workflow for co-authoring documentation. Use when user wants to write documentation, proposals, technical specs, decision docs, or similar structured content. This workflow helps users efficiently transfer context, refine content through iteration, and verify the doc works for readers. Trigger when user mentions writing docs, creating proposals, drafting specs, or similar documentation tasks. | Stars: 197k | Updated | Claude CodeCodex | Verified | Install |