josephgoksu/taskwing/taskwing-plan
Claude Code and agent skill by josephgoksu/taskwing
- community
Checking that the SKILL.md file is still in the repository.
Use when you need to clarify a goal and build an approved execution plan.
Install
npx vskill@latest install josephgoksu/taskwing/taskwing-planSKILL.md
Create Development Plan with Goal
Usage: /taskwing:plan <your goal> or /taskwing:plan --batch <your goal>
Example: /taskwing:plan Add Stripe billing integration
The Workflow Contract lives in CLAUDE.md (single source of truth). Obey it always.
Kill Table
| Impulse | Do Instead | |
|---|---|---|
| Auto-answer clarifying questions yourself | Present them to the user and WAIT | |
| Skip the clarification checkpoint | STOP with refusal text | |
| Plan without clarifying first | Always run clarify before generate | |
| Assume the user approved | Require explicit "approve", "yes", or equivalent |
Operating Principles
- Goal clarity first. Never generate a plan from a vague goal. Run clarify until
is_ready_to_planis true. - User approves at every gate. Checkpoints are hard gates, not suggestions. Missing approval = STOP.
- Auto-start after finalize. Once the plan is approved, immediately run
taskwing task next --auto-startto begin work. Do not wait for the user to say/taskwing:next.
Hard gate for this command:
- Do NOT generate, decompose, expand, or finalize a plan until the clarified goal checkpoint is explicitly approved.
- If approval is missing, STOP and respond with:
"REFUSAL: I can't move past planning yet. Clarification checkpoint is incomplete. Please approve the clarified goal first."
How taskwing plan works
All plan actions go through a single CLI verb:
The output is always JSON (a PlanToolResult). Parse the response and extract the fields you need (clarify_session_id, is_ready_to_plan, plan_id, phases, tasks, …).
Mode Selection
The plan tool supports two modes:
- Interactive (default): Staged workflow with checkpoints at phases and tasks
- Batch (--batch flag): Original all-at-once generation
Check if $ARGUMENTS contains --batch:
- If yes: Use batch mode (Steps 1-4)
- If no: Use interactive mode (Steps 1-8)
BATCH MODE (when --batch is used)
Step 0: Check for Goal
If `$ARGUMENTS` is empty or not provided: Ask the user: "What do you want to build? Please describe your goal." Wait for user response, then use that as the goal.
If `$ARGUMENTS` is provided: Use $ARGUMENTS as the goal and proceed to Step 1.
Step 1: Initial Clarification
Extract from the JSON result: clarify_session_id, questions, goal_summary, enriched_goal, is_ready_to_plan, context_used.
Step 2: Ask Clarifying Questions (Loop)
CRITICAL: Do NOT answer questions yourself. Present them to the user and WAIT.
If `is_ready_to_plan` is false: Present each question to the user exactly as returned. Wait for the user to respond.
If user says "auto" or "skip":
If user provides answers: Pipe a JSON params blob to stdin to keep multi-line answers safe:
Repeat until is_ready_to_plan is true.
Step 3: Clarification Checkpoint Approval (Hard Gate)
Before generating:
- present
enriched_goaland assumptions - ask for explicit approval ("approve", "yes", or equivalent)
If approval is not explicit, STOP and use the refusal text above.
Step 4: Generate Plan
Step 5: Present Plan Summary
Display the generated plan:
After presenting the plan summary, immediately start working on the first task.
Do NOT wait for the user to say /taskwing:next - the plan was just approved, start executing.
INTERACTIVE MODE (default when no --batch flag)
Step 1: Check for Goal (Same as Batch)
If `$ARGUMENTS` is empty or not provided: Ask the user: "What do you want to build? Please describe your goal." Wait for user response, then use that as the goal.
Step 2: Clarify Goal
CRITICAL: Present all clarifying questions to the user. Do NOT answer them yourself. Wait for the user to respond before proceeding.
If the user says "auto" or "skip", run clarify again with "auto_answer": true.
Loop until is_ready_to_plan is true. Save the clarify_session_id and enriched_goal for subsequent steps.
CHECKPOINT 1: User approves the enriched goal before proceeding. If approval is not explicit, STOP and use the refusal text above.
Step 3: Decompose into Phases
This returns 3-5 high-level phases. Present them to the user:
CHECKPOINT 2: Ask user to:
- Approve phases as-is
- Request regeneration with feedback
- Skip specific phases
Step 4: Expand Each Phase (Loop)
For each approved phase:
This returns 2-4 detailed tasks for the phase. Present them:
CHECKPOINT 3 (per phase): Ask user to:
- Approve tasks and continue to next phase
- Request regeneration with feedback
- Skip this phase
Repeat for each phase until all are expanded.
Step 5: Finalize Plan
After all phases are expanded:
Step 6: Present Final Summary
After presenting the plan summary, immediately start working on the first task.
Do NOT wait for the user to say /taskwing:next - the plan was just approved, start executing.
Security
- Security scanPassed, 100 of 100Pattern checks on the skill's files.
- Model reviewNot runA model review is scheduled for the next rescan.
Quality evaluation
- Task evaluationNot runNo evaluation yet. Run one in Skill Studio.
Source
- SKILL.md in the repositoryCheckingskills/taskwing-plan/SKILL.md
Popularity
- Repository stars
- 87
- Repository forks
- 6
- 7-day trend score
- 22
Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.
Works with
Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.
- Claude Code
- Cursor
- GitHub Copilot
- Windsurf
- Codex
- Gemini CLI
- Cline
- Amp
- Roo Code
- Goose
- Continue
- Kilo Code
- Junie
- Trae
- OpenHands
- Qwen Code
- AI Agent
- Tabnine Chat
- Sourcegraph Cody
- Amazon Q
- Firebase Genkit
- Sweep
- Devin
- Phind
- Mentat
- Claude Desktop
- Aider
- GPT Pilot
- Plandex
- bolt.new
- v0
- Replit Agent
- Anse
- TypingMind
- BetterChatGPT
- LibreChat
- Warp AI
- Pieces
- Fig AI
Embed badge
[](https://verified-skill.com/skills/josephgoksu/taskwing/taskwing-plan)