Task Management
Usage
/task-management or /task-management <action> <task-name>
Workflow Selection
Ask: "Will this change how a developer uses this package?"
If YES → Docs-First
- •Create task file pointing to MDX location(s)
- •Draft docs first (
website/docs/[package]/[feature].mdx) - •Iterate until docs are approved
- •Add implementation phases to task file (goals + success criteria, not steps)
- •Implement, create
test/for-docs/tests - •Replace
@skip-checkwith@check-example— use/doc-example-auditto get a categorized report, then convert interactively (see also CLAUDE.md Documentation section for annotation rules) - •Phase retro (see below)
- •Archive when complete
Key: MDX dominates. Task file holds implementation details, not user-facing API.
Especially valuable for:
- •Multi-component flows (browser ↔ server ↔ external services)
- •User-facing APIs where ergonomics matter
- •Integration points between systems
- •Anything involving redirects, callbacks, or multi-step handshakes
Docs-to-implementation handoff: When the docs phases are done and implementation begins (often in a new session with fresh context), the docs are a specification, not a description of existing reality. Code examples with @skip-check don't work yet — they describe the target API. The task file should clearly mark this transition, and session prompts should state: "The docs in website/docs/[package]/ are the spec — code examples describe the target API, not current behavior."
If NO → Implementation-First
- •Add to
tasks/backlog.md(small) or createtasks/[project-name].md(multi-phase) - •Define phases with goals + success criteria
- •Implement, updating task file as you go
- •Phase retro (see below)
- •Archive when complete
Better suited for:
- •Internal utilities and helpers
- •Algorithms where the interface is already clear
- •Refactoring with stable external API
- •Bug fixes
Task File Structure
See tasks/README.md for templates. Key elements:
- •Goal: One sentence
- •Phases: Each with goal + success criteria (not detailed steps)
- •Status: Design Complete | In Progress | Complete
Phase Retro
After each phase, briefly answer:
- •What did we learn? (surprising discoveries, undocumented behavior, patterns worth capturing)
- •What did we struggle with? (implementation friction, confusing APIs, wrong assumptions)
- •Did any tests fail unexpectedly? (root cause, not just the fix)
- •Impact on follow-on work? (does this change later phases, create new backlog items, or simplify/complicate the plan?)
Capture anything reusable (patterns, conventions, gotchas) in the appropriate place: CLAUDE.md, skill files, AGENTS.md, or backlog.md. Don't let hard-won knowledge stay only in the conversation transcript.
Rules
- •Ask "Ready to proceed with [next phase]?" after each phase
- •Run a phase retro after each phase completes
- •Update task file when plans change
- •Archive completed tasks (don't delete)