메뉴
Ultraplan cloud planning — kick off a plan in the cloud from your terminal, review and revise in the browser, then execute remotely or send back to CLI
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
Select and coordinate multi-agent teams (topology kits, role-based squads, lifecycle, worktree isolation). Use this skill whenever launching parallel agents, designing a review board, running a debug council, scheduling an orchestrator-workers team, configuring agent tool restrictions, or deciding between solo and team execution. Triggers on: "launch a team", "parallel agents", "review board", "debug council", "architect-implementer-reviewer", "swarm", "multi-agent", "subagents for X", "team topology", "agent lifecycle".
Select and wire an agentic design pattern (reflection, prompt chaining, routing, parallelization, orchestrator-workers, evaluator-optimizer, ReAct, blackboard) into the 5-layer Claude Code stack. Use this skill whenever deciding how to structure a multi-step task, whether to spawn subagents, how to run parallel review, or when to use which pattern. Triggers on: "which pattern", "orchestrate", "parallel review", "self-review", "chain of thought", "eval-optimizer loop", "blackboard", "ReAct", "how to decompose this task".
Auto mode permission handling — classifier-based approvals, PermissionDenied hook, defer permissionDecision, and autonomy profiles for hands-off Claude Code usage
Configure Claude Code's autonomous operating mode — profile selection (conservative, balanced, aggressive, unattended-review), permission blocks, and the three gate agents (planner, verifier, reviewer). Use this skill whenever enabling autonomous mode, switching profiles, tightening permissions for production branches, or setting up unattended execution. Triggers on: "autonomy", "unattended mode", "auto-approve", "permission mode", "autonomy profile", "gates", "/cc-autonomy", "planner verifier reviewer", "let claude run on its own".
Manage Claude Code's context window — token arithmetic, /compact strategy, anchor preservation, progressive loading, session analytics. Use this skill whenever a session gets long, context approaches limits, after /compact, when deciding what to load into CLAUDE.md vs leave in references, or when analyzing session cost/token usage. Triggers on: "context full", "compact", "too many tokens", "budget", "session analytics", "save tokens", "context window", "/compact strategy".
Evidence-driven deep analysis for hard coding problems — architecture decisions, root-cause investigation, high-stakes refactor planning, performance bottleneck isolation. Use this skill whenever the user asks for "the best approach", a "deep analysis", "root cause", "principal engineer review", or runs /cc-intel. Also triggers on hard debugging questions, major architectural choices, tricky performance problems, or any task where a hypothesis tree and evidence table matter more than a fast answer.
| name | cc-ultraplan |
| description | Ultraplan cloud planning — kick off a plan in the cloud from your terminal, review and revise in the browser, then execute remotely or send back to CLI |
| allowed-tools | ["Read","Bash"] |
| triggers | ["cc-ultraplan","/cc-ultraplan","cloud plan","plan mode cloud","remote plan","cloud planning"] |
Ultraplan offloads plan mode to the cloud. Claude drafts the plan in a Claude Code web session while your terminal stays free. You review sections in your browser, comment, request revisions, then choose to execute remotely or send it back to your CLI.
Available since v2.1.92 (research preview). As of v2.1.101, the first run auto-creates a default cloud environment — no web setup step required.
| Scenario | Use Ultraplan | Use Local Plan Mode |
|---|---|---|
| Complex multi-service migration | ✓ | |
| Need to share plan with teammates | ✓ | |
| Terminal must stay free during planning | ✓ | |
| Quick single-file refactor | ✓ | |
| No internet / air-gapped environment | ✓ | |
| Iterating rapidly on small tasks | ✓ |
/ultraplan migrate the auth service from sessions to JWTs
Include the keyword "ultraplan" in any prompt:
> ultraplan: design the database migration strategy for the multi-tenant switch
The browser interface shows the plan as structured sections:
Plan history is preserved — you can compare revisions and revert.
Claude Code on the web runs the implementation. Results and changed files are available in the session. Push from the web session or download a patch.
The approved plan is serialized and sent back to your local terminal session. Claude continues from there with the agreed plan as context.
After executing in the cloud and pushing a branch, use /autofix-pr to hand the PR over to Claude:
> /autofix-pr
Claude detects the open PR for your current branch and enables auto-fix — watches CI, fixes failures, handles review comments, pushes until green.
Ultraplan uses Opus (opus → latest claude-opus-4-8) by default for the planning phase — the reasoning depth justifies the cost for complex migrations and architectural decisions. Pair it with a high effort level (xhigh/max on Opus 4.8) for the hardest plans. For plans above Opus's ceiling — multi-day migrations, problems Opus has already failed on — escalate planning to Fable 5 (fable → claude-fable-5); state the goal and constraints rather than enumerating steps, since over-prescriptive scaffolding reduces Fable output quality. The execution phase uses the model appropriate to the task.
Estimated cost for a complex plan: ~$0.50–$2.00 depending on context volume. Simple plans use fewer tokens and run faster.
Use ultraplan to design a feature, then run /team-onboarding to generate a ramp-up guide that explains the resulting architecture to new teammates:
> /team-onboarding
This generates a guide from your local Claude Code usage patterns and the project structure — useful after landing a large feature.