一键导入
task-plan
Use only when the user explicitly requests this skill.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Use only when the user explicitly requests this skill.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
Do not use unless the user explicitly requests it. Dispatch each requested task to a new Codex session in a newly allocated dedicated worktree, then coordinate rebasing and fast-forward integration into local main without implementing, reviewing, or archiving the task in the coordinator session.
Use only when explicitly requested by the user or, during an explicit debugging session, after other relevant checks such as typecheck, lint, and tests have been exhausted.
Use when building greenfield React SPA UI with Lisse (`@lisse/react` or `@lisse/core`) for Figma-style smooth/squircle corners, choosing `SmoothCorners` vs `useSmoothCorners`, configuring corner smoothing, borders, gradient borders, shadows, styling hooks, auto-effects, client-side path generation, performance, or limitations.
Use React Aria and React Stately hooks to build accessible custom React components, including choosing the right hook, wiring refs/state/props, preserving ARIA and keyboard behavior, excluding React Aria i18n and SSR utilities, and checking hook-specific guidance in bundled category docs.
Design, review, and optimize PostgreSQL and SQLite schemas, queries, migrations, transactions, indexes, and data-access code, especially when SQL should remain portable between both engines.
Use when building React animations with Motion primitives, including motion components, animation props, gestures, layout/scroll/SVG animation, transitions, variants, Motion values, core React Motion hooks/components, and occasional vanilla JavaScript Motion APIs when React APIs are less appropriate or less capable.
| name | task-plan |
| description | Use only when the user explicitly requests this skill. |
| disable-model-invocation | true |
Use TASK.md as a temporary, git-ignored planning file for implementation work.
TASK.md is a pre-implementation contract between the agent and its supervisor. It records the approved task: outcome, relevant current behavior, intended behavior, scope, implementation shape, and commit plan.
It is not a scratchpad, log, checklist, transcript, or retrospective.
TASK.md at the repository root before editing code.TASK.md; it is temporary and already git-ignored.TASK.md.TASK.md. Preserve the resulting requirement, not the negotiation transcript.TASK.md as locked.TASK.md only when the supervisor changes requirements or explicitly authorizes restructuring.Write the smallest contract that preserves the supervisor-approved intent.
Include a detail only when omitting it would materially increase the risk of misunderstanding, scope drift, incorrect implementation, incorrect review, or incorrect commit grouping.
For design-driven work, include the full set of supervisor-approved design requirements that affect implementation, review, or acceptance. Preserve required layout, interaction, visual, responsive, accessibility, copy, and asset constraints unless they are clearly out of scope.
Do not include details merely because they are true, inferable, recently discovered, or potentially useful.
Prefer observable behavior and durable constraints over implementation narration.
Unless the task is mechanical, describe the implementation shape conceptually. Prefer packages, subsystems, APIs, commands, user flows, contracts, ownership boundaries, and externally visible behavior over exact files, modules, functions, classes, or symbols.
Use exact implementation references only when they are inherent to the task, supplied by the supervisor, or necessary to prevent misunderstanding.
Mechanical tasks are the exception: renames, moves, codemods, dependency bumps, config changes, generated artifacts, and similarly exact work may name precise paths, symbols, commands, and operations.
Before writing or finalizing TASK.md, identify whether ambiguity remains.
High-risk ambiguity is any unresolved choice where a wrong assumption could materially change behavior, scope, compatibility, data safety, security, architecture, ownership, reviewability, or commit structure.
Low-risk ambiguity is uncertainty the agent can safely resolve from local conventions, nearby code, tests, style, or the most likely interpretation without materially changing the task contract.
Surface ambiguity in conversation:
A., B., or C.. Reusing letters in a later pass of questions is fine.Only after high-risk ambiguity is resolved and the supervisor approves proceeding should TASK.md become the locked task contract.
Before requesting approval, reconcile the complete task conversation against TASK.md. Ensure the contract preserves all resolved requirements and constraints; if a clarified detail is intentionally omitted, confirm that its omission cannot affect implementation, scope, acceptance, review, or commit grouping.
Use the approval handoff to make the supervisor’s required attention explicit.
After creating or updating TASK.md, respond in conversation with an approval handoff.
The handoff is not a replacement for TASK.md. Its purpose is to route the supervisor’s attention.
State:
If high-risk ambiguity remains, do not ask for implementation approval. Ask 1–3 questions. Each question must include 2–3 plausible answers ordered by likely correctness. Prefix each suggested answer with a distinct capital letter within the same response.
If no high-risk ambiguity remains but the task is broad, architectural, risky, or expensive to unwind, recommend reviewing TASK.md before approval and identify the parts most worth checking.
If no high-risk ambiguity remains and the task is narrow, summarize the contract briefly and ask for approval to proceed.
The commit plan must contain at least one commit.
Use the fewest commits that preserve meaningful review and rollback boundaries. Additional commits are justified only when they isolate distinct behavior, tests, contracts, migrations, generated artifacts, documentation, mechanical refactors, or risk.
Do not split commits merely by chronology, package, implementation step, or because the task touches multiple areas.
Each commit should state its title, purpose, and likely area of work. Keep these conceptual unless exact references are necessary.
Once implementation begins, do not edit TASK.md to reconcile it with discoveries, changed tactics, specific files touched, verification details, progress, completed work, or retrospective corrections.
If implementation materially diverges from the locked contract, stop and ask the supervisor whether to restructure.
If divergence is immaterial, continue and disclose it in the final response.
Use this shape by default:
# Task
## Goal
State the requested outcome in one or two sentences.
## Current Behavior
Describe only the existing behavior needed to understand the change.
## Intended Behavior
Specify the desired behavior in observable terms, including relevant design requirements, constraints, edge cases, compatibility expectations, and non-goals.
## Scope
Define in-scope and out-of-scope boundaries when doing so prevents misunderstanding.
## Implementation Plan
Describe the implementation shape conceptually. Do not write a step-by-step migration screenplay.
## Commit Plan
List one or more reviewable commits.
For each commit, include title, purpose, and likely area of work.
Omit sections that would be empty or redundant.
Omit anything that does not preserve the approved contract, including:
Before committing or giving the final response, compare the work against the locked TASK.md.
Confirm that the work matches the approved contract, the commit grouping follows the approved plan, any material divergence was approved, and TASK.md is not staged.
Useful checks:
git status --short
git diff --cached --name-only