feature-tasks（特色任务）

Feature Tasks

Create planning artifacts in planning/<dd-mm-yyyy-title-slug>/ and generate task.graph.json.

Workflow

1. Clarify feature scope, constraints, and acceptance criteria.
2. Pick a title slug (for example add-new-payment-method). The generator resolves it to planning/<dd-mm-yyyy-title-slug>/ for new folders, while existing non-prefixed folders are reused.
3. Author planning/<dd-mm-yyyy-title-slug>/SPEC.md with: feature overview, requirements, constraints, and design decisions. Required by the companion orchestration skill.
4. Author planning/<dd-mm-yyyy-title-slug>/tasks.yaml using the strict schema below.
5. Keep related planning artifacts in the same folder (SPEC.md, task.graph.json, task.status.json).
6. Run the bundled script to validate and generate task.graph.json.
7. Fix all schema and parity errors until generation succeeds.
8. Use task.graph.json to drive execution order.

tasks.yaml schema (strict)

version must be 3. Version 3 supports the optional context field per task.

version: 3
project: <project-name>
tasks:
  - id: TASK-001
    phase: planning
    priority: high # low | medium | high | critical
    title: Define acceptance criteria and scope
    blocked_by: []
    acceptance:
      - Acceptance criteria reviewed by stakeholders
    deliverables:
      - docs/spec.md updated with final scope
    owner_type: docs # frontend | backend | infra | docs | qa | fullstack
    estimate: M
    notes: Optional context
    context:
      - src/api/orders.ts
      - SPEC.md#payment-flow
      - Uses Repository pattern from src/lib/repo.ts
critical_paths:
  - id: cp-main
    tasks: [TASK-001]
parallel_windows:
  - id: pw-initial
    tasks: [TASK-001]

Required task fields

Every task must include:

• id (unique string)
• phase (string)
• priority (low|medium|high|critical)
• title (string)
• blocked_by (array of task IDs; empty allowed)
• acceptance (non-empty array of strings)
• deliverables (non-empty array of strings)
• owner_type (frontend|backend|infra|docs|qa|fullstack)
• estimate (string)

Optional:

• notes (string)
• context (array of strings): File paths, spec section refs, or hints pointing worker agents to relevant code. Example: ["src/api/orders.ts", "SPEC.md#payment-flow", "Uses Repository pattern from src/lib/repo.ts"]

Task granularity

• Each task should be completable in a single agent session (typically 1-3 files changed).
• If acceptance criteria span multiple subsystems, split into separate tasks with dependencies.
• Prefer more small tasks over fewer large ones — parallelism improves throughput.
• A task that changes more than 5 files is usually too broad.

Execution

Run from the target repository:

<path-to-skill>/scripts/generate-task-graph [title-slug] [output-graph-json]

Default behavior:

• Input: <git-root>/planning/<dd-mm-yyyy-title-slug>/tasks.yaml
• Output: <git-root>/planning/<dd-mm-yyyy-title-slug>/task.graph.json
• If title-slug is omitted, the script derives it from the current branch name, strips common conventional prefixes (feat-, fix-, etc.), and resolves new folders to a dd-mm-yyyy-<slug> prefix (existing non-prefixed folders are reused).

Validation and parity checks

The script fails with actionable errors when:

• Required fields are missing
• version is not 3
• priority or owner_type enum values are invalid
• Task IDs are duplicated
• blocked_by references unknown IDs
• critical_paths[*].tasks or parallel_windows[*].tasks reference unknown IDs
• A task depends on itself
• Dependency cycles exist

Companion Skill

Use feature-tasks-work for orchestration mode after planning is complete.