fuzhiyu superplan — the PLAN phase
First, load superRA:using-superra and superRA:communicate if not already loaded.
Harness plan mode active: load references/harness-plan-mode.md before proceeding.
Overview
Output: a superRA/ task tree to execute.
Task-tree design judgment — objective/guidance writing, placement, splitting, context distillation, update-task lifecycle, retroactive creation — lives in references/task-tree-design.md.
Announce at start: "I'm using the superplan skill to create the task tree."
Output location: superRA/ at the worktree root if in a worktree, otherwise the project root, unless the user specifies elsewhere. Commit the task tree before execution.
Entry Assessment
Assess three dimensions before exploration or task design — creating a tree and updating one both pass through this.
- Placement.
superRA/exists: place byreferences/task-tree-design.md§Placing Work in the Existing Tree. LegacyPLAN.mdwithout a tree: offer migration (task-tree§Migration). Neither: the work becomes the first top-level task undersuperRA/. - Depth tier. Choose from §Depth Tiers.
- Routing path. Forward planning (default), or retroactive documentation — code/results without task coverage need a
superRA/record; same phases, pertask-tree-design.md§Retroactive Task-Tree Creation. Structural cleanup of an existing tree is neither — that is the separatereferences/consolidation.mdpass, entered on structural debt, not on new work needing placement.
Interactive canvas cadence (the default, and where light planning and execution move together): load using-superra/references/interactive-mode.md.
Ask when unclear. Placement or depth ambiguous, or the request too vague to aim exploration: run a scoping round per §Grilling — candidate placements from the descent, standard vs. thorough. Don't guess.
Depth Tiers
A spectrum, not rigid modes — escalate mid-planning when complexity warrants. The tier mainly modulates Phase 1; Phase 4 is identical except thorough adds §Agent Review.
| Tier | Use for | Phase 1 |
|---|---|---|
| Quick | Minor updates, known additions, single-task changes — an objective rewrite after a scope revision, a well-understood subtask, a dependency adjustment. | Light scan of superRA/, skip deep exploration, design inline. |
| Standard (default) | New workstreams in familiar territory, a significant new branch, work a domain skill governs. | Explore project structure, load the domain skill, design tasks. |
| Thorough | Complex or unfamiliar projects, large scope across codebase areas, or an explicit request ("plan hard", "explore thoroughly", "detailed plan"). | Dispatch parallel exploration agents per references/thorough-planning.md; competing designs feed §Grilling. |
Phase 1: Exploration
Read project structure, existing code, data directories, documentation, CLAUDE.md/README.md, and git history for relevant prior work, scaled to the tier (table above). Domain planning-survey gathering (data inventory, model primitives survey, manuscript assessment) begins here.
Exploration shifts placement or depth: revisit the entry assessment.
Phase 2: Domain Setup & Scope
Stop here, load the matching domain skill and follow its planning-stage reference per its own stage-load table. Run its planning survey before any task structure is drafted: the survey lands in ## Details, and the decisions it raises join the frontier for the Phase 3 round.
No implemented domain skill for the work: proceed to Phase 3, flag the gap to the researcher.
Phase 3: Design & Task Decomposition
Grill before decomposing. Run the main round per §Grilling.
Then load references/build-and-review.md; follow its procedure.
Phase 4: Review & Commit
Self-Review
Run references/build-and-review.md §Self-Review.
Agent Review
At thorough depth, dispatch Stage: planning-review before presenting the tree to the user. Explicit handoff-review requests enter the same step. Load superRA:agent-orchestration before writing the dispatch prompt.
Planning reviewer:
Prompt:
Load `superRA:review-task` skill.
Stage: planning-review
Task: <task path or root>
Review mode: handoff-readiness | design-review
Context: <exploration synthesis, inline or path>
Review modes: references/planning-review.md §Review Mode — design-review for a newly authored tree, handoff-readiness once the design is settled.
Fix REVISE findings before User Review.
Quick and standard depth: skip this step unless the user asks for a handoff review.
User Review
Present the tree (superra task tree) with the dashboard link. A tradeoff the assembled tree exposes as still unsettled re-enters §Grilling.
Per newly created task (especially top-level): state the existing concern you considered and why it does not cover the work.
Execution Handoff
Commit superRA/ atomically (plan(add): <summary> for initial authoring; full sub-step set in §User Feedback and Changing the Task Tree). Then execute the frontier in the current execution mode (superRA:using-superra/references/main-agent.md §Execution Modes).
Grilling
Put every unsettled decision to the researcher in frontier-ordered rounds, each question carrying your recommended answer: references/grilling.md. Standard and thorough depth grill by default; quick depth skips it unless the researcher asks.
User Feedback and Changing the Task Tree
Route task-tree changes through references/changing-the-tree.md.