Agent-Skills.md

Agent Skills: Writing Tasks

Decomposes specs into persistent task files under tasks/ with a dependency graph, parallel lane suggestions, and dynamically derived progress. Use when user mentions task 분리, task 생성, task 쪼개기, spec을 task로, decompose, writing-tasks, 병렬 작업 편성, 의존성 그래프. Proactively trigger when a specs/ directory has unmapped specs or when user wants to track long-running work as persistent task files.

UncategorizedID: devstefancho/claude-plugins/writing-tasks

Repository

devstefancho/claude-plugins
devstefancho
111

Install this agent skill to your local

pnpm dlx add-skill https://github.com/devstefancho/skills/tree/HEAD/skills/spec-driven/writing-tasks

Skill Files

Browse the full folder contents for writing-tasks.

Download Skill

Loading file tree…

skills/spec-driven/writing-tasks/SKILL.md

Skill Metadata

Name
writing-tasks
Description
Decomposes specs into persistent task files under tasks/ with a dependency graph, parallel lane suggestions, and dynamically derived progress. Use when user mentions task 분리, task 생성, task 쪼개기, spec을 task로, decompose, writing-tasks, 병렬 작업 편성, 의존성 그래프. Proactively trigger when a specs/ directory has unmapped specs or when user wants to track long-running work as persistent task files.

Writing Tasks

Decompose specs from specs/ into one-file-per-task under tasks/. Progress, lanes, and graphs are always computed from task frontmatter — never stored.

Hard Rules

  • Task files are the only SSOT. Never create tasks/README.md, tasks/DEPENDENCIES.md, tasks/TEMPLATE.md, or status scripts in the user's repo — everything is derivable.
  • depends_on is mandatory. An empty array must be consciously declared, never defaulted.
  • blocks ↔ depends_on never drift. When A.depends_on gains B, add A to B.blocks automatically (write-through invariant, silent auto-fix).
  • Validate on every write. Cycles, duplicate ids, phantom deps/specs are rejected before the file lands. Full rules in task-schema.md.
  • Parallel-worktree safe. One task per file; no shared files under tasks/ means worktrees never collide.
  • Spec-driven by default. No-arg invocation decomposes specs/; manual new is the exception.

Layout (v1 — fixed)

specs/  phase-N/NN-slug.md            ← input (user-authored or writing-specs)
tasks/  phase-N-slug/NN-task-slug.md  ← one task per file
  • Phase numbers increase monotonically across the project lifetime; never reset.
  • Phase directory slug is required, human-readable (phase-1-foundation).
  • Local task number is 2-digit per phase; the global id is N.NN (e.g. 3.02 = phase 3, task 2). Cross-phase references always use N.NN.

Frontmatter schema and validation rules: task-schema.md. Task body sections come from templates/task-template.md.

Commands

Only two user-facing commands. Everything else is derived.

/writing-tasks (no args) — smart dispatch

| State | Action | |---|---| | Specs without matching tasks | Propose decomposition, await approval, create files | | All mapped, work in progress | Print the status dashboard (dashboard.md) | | New specs/phase-N/ with no tasks/phase-N-*/ | Ask for the phase slug, then decompose | | All tasks done | Print All done. + one-line suggestion (next phase or archive hint) |

/writing-tasks new <description> — manual task

For ad-hoc tasks without a spec (hotfixes, refactors):

  1. Ask which phase (multi-select existing phases + + new phase).
  2. Show existing tasks (id, title, status); ask user to select depends_on (multi-select).
  3. None selected → re-confirm "Is this truly independent (no prerequisites)?" — explicit opt-in required.
  4. Ask estimate (S/M/L), generate next local number, write from templates/task-template.md.
  5. Auto-sync blocks on each dependency, then validate.

Decomposition (primary path)

Sizing rules, the four dependency-inference signals, preview format, and report format: decomposition.md.

  1. Scan — Glob specs/**/*.md + Glob tasks/**/*.md; diff specs not referenced by any task's spec: field.
  2. Propose — split each unmapped spec into tasks, infer depends_on from the four signals, show the preview table with the inference trail (which signal → which dependency) so the user can correct.
  3. Gate — AskUserQuestion: proceed / edit / cancel. On edit, modify dependencies interactively. Never write before this gate.
  4. Write — create tasks/phase-N-slug/ if new (ask for slug), write each task from the template, validate once after all writes.
  5. Report — phases created, task count, cross-phase deps documented, validation result.
  • [ ] Preview approved (or non-interactive default applied and echoed in the report)
  • [ ] Validation clean after all writes

Incremental updates (re-runs)

  • New spec → add new task(s); never overwrite existing files.
  • Modified spec → never silently rewrite the task. Ask: "Spec X was modified. Resync task N.NN?" with a diff.
  • Deleted spec → mark the matching task as orphan (warning); never auto-delete.
  • status: done task → never modify its body on resync unless the user explicitly asks.

Worktree integration

Inside a .claude/worktrees/task-N.NN-* worktree: parse the branch name to detect the task, offer the todo → in_progress transition, and show only that task's checklist as "Next up" in status output.

Anti-patterns

WRONG: write tasks/README.md or a scripts/task-status.ts to track progress. RIGHT: compute the dashboard from task frontmatter on demand (dashboard.md).

WRONG: quietly default depends_on: [] when no dependency is obvious. RIGHT: require explicit independence — user opt-in, or the documented non-interactive default in decomposition.md.

Boundaries

This skill only writes and validates task files. Specs come from writing-specs (the natural pair); implementation, tests, and status flips flow through implement-with-test. Worktree management and personal todo lists are out of scope. Keep outputs tight — the dashboard fits on one screen.