メニュー
Generate a tasks.md for a speckit feature. Explicit invocation only — never load from context or topic. Use only when the user types the exact command "speckit-tasks".
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
| name | speckit-tasks |
| description | Generate a tasks.md for a speckit feature. Explicit invocation only — never load from context or topic. Use only when the user types the exact command "speckit-tasks". |
| disable-model-invocation | true |
Generates an actionable, dependency-ordered tasks.md from the feature's plan and spec artifacts.
Run: .specify/scripts/bash/check-prerequisites.sh --json from repo root. Parse FEATURE_DIR and AVAILABLE_DOCS.
plan.md (tech stack, architecture), spec.md (user stories, requirements, content model) and draft test contentdata-model.md, research.md, quickstart.md.specify/templates/tasks-template.md as structure templateSimple (2-4 tasks): CSS-only tweaks, minor JS fixes to an existing block, single-file changes.
Standard (5-8 tasks): New block (JS + CSS), auto-blocking changes, block with multiple variants.
Complex (9-15 tasks): Multiple new blocks, core script changes combined with new blocks, third-party integrations via delayed.js, cross-block interactions.
State complexity in tasks.md header.
- [ ] [Category-ID] [P] Description with file path
| Category | Prefix | Typical files |
|---|---|---|
| Block JS | BJ | blocks/{name}/{name}.js |
| Block CSS | BC | blocks/{name}/{name}.css |
| Core Scripts | CS | scripts/scripts.js, scripts/delayed.js |
| Global Styles | GS | styles/styles.css, styles/lazy-styles.css |
| Content | CT | drafts/ test content, head.html, icons |
| Integration | IN | Third-party scripts, auto-blocking, indexing |
| Testing | TS | Linting, browser tests, PSI validation |
| Documentation | DC | Block README, authoring guides |
| Setup | T | Project structure, configuration |
| Scaffolding | SC | New block directory creation |
Include [P] only when task operates on different files than all incomplete tasks.
| Collapse | To | Exception |
|---|---|---|
| SC001+BJ001+BC001 | "Create block with JS and CSS" | Keep separate if block has complex decoration AND complex styling |
| BJ001+BC001 | "Implement block decoration and styles" | Keep separate when design.md exists (CSS from design, JS from content model) |
Never skip: SC001 (new block), BJ001 (block JS for any block with decoration logic), TS001+TS002 (always — linting gates). TS006 when logic-heavy code is present per Logic-Heavy Block Detection.
Create the block directory and files:
blocks/{blockname}/
├── {blockname}.js # exports default function decorate(block)
└── {blockname}.css # scoped to .{blockname}
Use the building-blocks skill for the decorate function pattern and CSS conventions.
T001 verifies that test content exists, uses the correct file format, and renders decorated blocks in the browser. This is a gate — do not proceed to implementation until T001 passes.
T001 must include ALL of the following checks:
.plain.html extension per the HTML Structure Guide. Files must contain only section content (<div> wrappers with blocks/default content) — NOT a full HTML page with <!DOCTYPE html>, <head>, or <body>.aem up --html-folder . (or --html-folder drafts if files are in drafts/). Without this flag, .plain.html files will 404.localhost:3000/drafts/block-test for drafts/block-test.plain.html) and confirm blocks are decorated — the AEM decoration pipeline (scripts.js) must run and the block's decorate() function must execute. Raw content with no styling = broken.Common failure mode: Content file created as .html (standalone page) or .plain.html accessed directly at the .plain.html URL — both bypass the AEM decoration pipeline and serve raw undecorated HTML. The correct URL never includes the file extension.
When design.md exists:
@media (width >= 600px) → 900px → optional 1200px). Match ## Layout matrix (flex / grid) for flex-direction / gap per breakpoint; repeat desktop overrides when they differ from tablet.styles/styles.cssCT001 verifies that all block variants render correctly in the browser after implementation. This is NOT a file-existence check — it requires loading the page and confirming visual/functional correctness.
CT001 must specify:
localhost:3000/drafts/block-test)CT001 is NOT done until someone (human or automation) has loaded the URL in a browser and confirmed the blocks are decorated and functional. Checking file contents or DOM structure in code is not a substitute for browser verification.
npm run lint must pass before PRblocks/{blockname}/{blockname}.test.js for isolatable functions. Follow testing-blocks and block-unit-testing resources. Skip for decorative/wiring-only blocks.Add TS006 when the block's BJ001 description or data-model/research indicate:
Do NOT add TS006 for: pure DOM reshaping (rows → list), simple class toggling, wiring-only decorate().
Output: tasks.md path, task count, parallel opportunities, format validation confirmation, and readiness for next phase: /speckit-implement (default when tasks are complete), or optionally /speckit-analyze (to inspect task structure before implementation).
Manual asset processing skill for Figma layers. Accepts layer URLs and writes deterministic files to drafts/media using curl+sips, with optional get_screenshot fallback.
Phase 3 of athenai-implement pipeline — execute tasks from plan.md and produce working block code. Includes lint-aware CSS generation and mandatory design compliance check when design.md exists.
Phase 0 of athenai-implement pipeline — conditionally produce design.md from Figma. Includes decision logic, extraction-only rules, manual asset handoff, and Phase 1 context pre-loading optimization.
End-to-end implementation pipeline — design, plan, tasks, implement, test cases, test content. Demo mode: skips unit tests and research deep-dives. Records phase wall times via `.specify/scripts/phase-timer.py` (per-phase start/end, subagent logs + merge when parallel). Explicit invocation only — never load from context or topic. Optional flags: --no-parallel (run all phases sequentially in main agent, no subagents), --no-test-cases (skip Phase 4 test case generation).
Phase 1-2 of athenai-implement pipeline — produce plan.md with authored HTML structure, parsing logic, and tasks section. Includes draft test content generation with mandatory edge case coverage.
Phase 4 of athenai-implement pipeline — generate testcases.csv from spec.md. Includes guard for --no-test-cases flag and coverage requirements for user journeys, acceptance criteria, and edge cases.