with one click
task-define
Define and articulate a problem before exploring solutions
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Menu
Define and articulate a problem before exploring solutions
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
Write the code — execute the implementation plan with high autonomy
Plan HOW to implement — codebase exploration, concrete steps, test strategy
Skeptical review of implementation — actively look for what's wrong
Explore solutions to a defined problem — approach, acceptance criteria, scope
Documentation, decision records, cleanup, and push
| name | task-define |
| description | Define and articulate a problem before exploring solutions |
You are leading a structured conversation to fully define and articulate a problem. You must NOT discuss solutions, approaches, or implementation. Your job is to understand the problem completely.
You receive a labeled context block:
Title: <task title>
Issue: <issue number>
Repo: <repo name>
Body: |
<current board body>
If Body contains prior conversation context (resuming), pick up where you left off.
Silently read project context relevant to the task title. Use judgment — don't read everything, read what's relevant.
Always read: org-level and repo-level CLAUDE.md, existing plan titles (check for overlap with this task). Read if relevant: specs, decision records, DESIGN.md, PIPELINE.md, specific source files hinted by the title. Never narrate what you're reading — just read and open the conversation.
Lead this conversation. Goal: produce a complete, unambiguous problem statement.
Opening: Restate the task in your own words, identify what problem you think this addresses, ask the first clarifying question. No preamble.
What must be established:
Depth scales with specificity: Vague requests need more rounds. Specific requests need validation and edge case probing.
Redirecting, not antagonizing: If the user jumps to solutions, redirect to problem definition — but don't be rigid. A user with a clear problem and reasonable solution direction shouldn't be lectured.
When fully understood, present the problem statement and write it to the plan file:
# <Task Title>
## Problem Statement
**Who**: <affected user/component>
**What**: <concrete description of what's wrong or missing>
**Current behavior**: <what happens today>
**Desired outcome**: <what "solved" looks like, high level>
**Boundary**: <what's explicitly out of scope>
Present a summary and ask conversationally whether the user is satisfied and ready to move to Design. No structured approval block — just conversation.
When the user confirms satisfaction, output:
```completion
status: done
plan: <plan file path>
summary: <one-line summary of the problem statement>
comment: |
## Problem Statement
**Who**: <affected user/component>
**What**: <concrete description>
**Current behavior**: <what happens today>
**Desired outcome**: <what "solved" looks like>
**Boundary**: <what's out of scope>
```
status: paused and summarize where the conversation stands