task-planner
When decomposing a feature or spec into implementable subtasks, estimating scope, or orchestrating multi-step work across subagents.
- License 许可
- MIT
- 最近更新
- 2026-11-07
- Stars
- ★ 13
- Forks
- 3
01 前言 Front Matter
- name
- task-planner
- license
- MIT
- description
- When decomposing a feature or spec into implementable subtasks, estimating scope, or orchestrating multi-step work across subagents.
03 SKILL.md 正文
Task Planner Standards
Task Management
- Plan First: Write plan with checkable items before starting.
- Verify Plan: Check in with the user before executing.
- Track Progress: Mark items complete as you go.
- Explain Changes: High-level summary at each step.
- Document Results: Add review section when complete.
- Capture Lessons: Update
.memory/LESSONS.mdafter corrections.
Task Quality
- Tasks must be clear, specific, and actionable.
- Complex tasks must be broken into atomic, trackable subtasks.
- Dependencies between tasks must be explicitly declared.
- Security-related tasks must undergo mandatory review.
- Escalate ambiguous or contradictory tasks for clarification.
Blueprint Check (if configured)
If Blueprint Path is set in .memory/CONFIG.md and the task involves shared
business logic:
- [ ] Read the relevant spec(s) in
[BLUEPRINT_PATH]/specs/before planning. - [ ] Check
[BLUEPRINT_PATH]/parity/PARITY_MATRIX.mdfor current platform status. - [ ] Include in the plan: "Requires Knowledge Bus entry" if the change affects cross-platform contracts.
- [ ] Note any spec gaps — if the spec is missing or incomplete, plan a spec update as part of the task.
Skip this section if no blueprint is configured or the task is project-specific.
Change Tracking
- All rule or config changes must be documented in CHANGELOG.md.
- Record source, timestamp, and rationale when modifying shared assets.
- A rollback plan must be defined for every major change.
Subagent Strategy
When a task benefits from fresh context, parallel execution, or isolated scope,
use the subagent-dispatch skill. It provides prompt templates for
implementer, reviewer, and researcher subagents with explicit scope boundaries
and result validation.
When to dispatch:
- Subtasks are independent and can proceed in parallel
- Parallel execution provides meaningful time savings
- Fresh context helps (no accumulated conversation noise)
- Scope of each subtask is clear and self-contained
When NOT to dispatch:
- Iterative feedback from the user is needed between steps
- The task is simpler to do directly in the current session
- Subtasks need full conversation history or shared state
- Scope is unclear or likely to shift during execution
04 同仓库更多 Skills
setup-project
When setting up agentskel on a new project that does not yet have .memory/ or .agents/. Use when the user says "set up agentskel", "install agentskel", or "add agent memory to this project".
using-git-worktrees
When running long feature implementations, parallel branch work, or build/test scripts that must not interfere with the user's active working directory. Creates isolated execution environments via git worktrees in sibling directories.
test-engineer
When designing test strategy, analysing coverage, or writing tests outside the TDD cycle. Also use when working with CI pipelines or validating changes end-to-end.
test-driven-development
When implementing logic changes during develop-feature or implement-task. Use the Red-Green-Refactor cycle to write a failing test before any production code.
task-planner
When decomposing a feature or spec into implementable subtasks, estimating scope, or orchestrating multi-step work across subagents.
systematic-debugger
When applying debugging techniques during any investigation — root cause tracing, bisect, defense-in-depth. Enforces hypothesis-first analysis and prohibits guess-and-check changes. Use standalone or inside the debug-issue workflow.