Stats
Actions
Tags
From flow
Run the full flow workflow end-to-end — prep → optional research → plan (with optional multi-LLM brainstorming) → optional plan-review → develop → deploy — based on a single task goal from the user. Use this when the user wants to hand off a complete task and let the workflow run, rather than driving each step manually. Skips research, brainstorming, and plan-review automatically for size S tasks; runs the full pipeline for size L. Even when the user just says "build me X", consider this skill if the task warrants the full discipline.
How this skill is triggered — by the user, by Claude, or both
Slash command
/flow:orchestrateThe summary Claude sees in its skill listing — used to decide when to auto-load this skill
Orchestrate is a thin sequencer. It does not reimplement any of the individual skills — it invokes them in order, with skip logic based on size and explicit user signals.
Orchestrate is a thin sequencer. It does not reimplement any of the individual skills — it invokes them in order, with skip logic based on size and explicit user signals.
1. flow:prep
└── creates worktree, branch, .planning/, prepare.md (with size estimate)
2. flow:research [skip if size = S, or user opted out]
└── writes research.md
3. flow:plan [always]
└── (sub-phase) multi-LLM brainstorm if size = L, or size = M with cross-module /
security-sensitive / public-surface flag → writes brainstorm.md + artifacts/brainstorm-*.md
└── writes plan.md, tasks.md
4. flow:plan-review [run if size = L; ask user if size = M; skip if size = S]
└── writes artifacts/plan-review-*.md and artifacts/plan-review-summary.md
└── if substantive changes: bumps plan.md → artifacts/plan.v1.md, writes new plan.md
5. — Checkpoint with user —
Show plan.md, tasks.md, and plan-review-summary.md (if exists). Wait for go/no-go.
6. flow:develop [after user confirms]
└── executes tasks.md, atomic commits, all checkboxes filled
7. flow:deploy [as a separate session — see below]
└── pushes, opens Korean PR, runs multi-LLM review, posts inline comments
| Step | size = S | size = M | size = L |
|---|---|---|---|
| prep | yes | yes | yes |
| research | skip | ask | yes |
| plan (always) | yes | yes | yes |
| ↳ brainstorm sub-phase | skip | ask (default yes if cross-module / security / public-surface) | yes |
| plan-review | skip | ask | yes |
| user checkpoint | skip | yes | yes |
| develop | yes | yes | yes |
| deploy | yes | yes | yes |
"Ask" means: surface the decision to the user with the size-based default pre-selected. Don't bounce every step.
This is the only mandatory pause in orchestrate. Show the user:
artifacts/plan.ko.md, or artifacts/plan-review-summary.md if plan-review ran)Wait for an explicit go-ahead before invoking flow:develop. The reason for the pause: develop runs for a while and produces commits — the user should sign off on what is about to be built. After the checkpoint, develop runs without further interruption unless it hits a blocker.
Deploy intentionally runs as its own session. Orchestrate's job at the end of develop is:
tasks.md is fully checked.flow:deploy to open the PR and run the multi-LLM review."Do not auto-invoke deploy inside orchestrate. The reasons:
If the user objects and explicitly says "just run deploy too", you may invoke it inline, but mention the trade-off.
Each sub-skill should report its outcome. If any step fails:
../../references/multi-llm.md): ≥2 valid reviews → continue with synthesis and a ## 결손 리뷰어 note; 1 valid → ask user; 0 valid → stop. Orchestrate does not override these decisions.flow:develop directly when they want to continue.Do not retry silently. Orchestrate is a sequencer, not a self-healing pipeline.
Each individual skill is the source of truth for its own behavior. This skill only sequences them:
flow:prepflow:researchflow:planflow:plan-reviewflow:developflow:deploynpx claudepluginhub gitgitwi/council-flow --plugin flowGuides creation, editing, and verification of skills for AI coding agents using test-driven development with subagent scenarios. Use when authoring or debugging skills.