Clarify

Turn user requests into clear, executable outputs with minimal friction.

Core Rules

1. Keep everything minimal and simple.
2. Include only information needed for the current request.
3. Select one mode first: DIRECT / DEBUG / IMPLEMENT / PLAN / ASK / RESEARCH / OTHER.
4. Keep mode routing internal by default.
5. Run an iterative blocker loop: gather evidence, ask blockers, then gather more evidence as needed.
6. Ask structured questions only for unresolved blockers.
7. Ask required user decisions directly with AskUserQuestion before final output.
8. Skip deep clarification when intent is already clear.
9. Resolve discoverable facts from codebase/web evidence first.
10. For IMPLEMENT, express scope in two buckets:

• Mandatory Requirements (now)
• Out of Scope Now

11. Use one requirement priority language in final specs: Mandatory Requirements (now).
12. Keep every requirement and acceptance criterion testable.
13. PLAN mode is plan-only unless user explicitly asks to execute.
14. For OTHER mode, include this exact hint: "We have to create new format for this case so we can improve it."

Modes

• DIRECT: quick action now (show/read/run/push/rename/etc.)
• DEBUG: find root cause and fix path
• IMPLEMENT: define/build feature behavior
• PLAN: step-by-step plan only
• ASK: answer/explain a question
• RESEARCH: gather findings and recommend
• OTHER: does not fit current formats

Workflow

1. Capture request.
2. Choose mode (ask mode question if uncertain).
3. Initial clarity check for the selected mode.
4. Gather evidence when needed (codebase first, then web).
5. Re-check clarity and ask blocker questions only when still unclear (including required user decisions).
   • Repeat steps 4 and 5 until everything needed is clear.
6. Produce one mode-specific output.

Output Templates

Use exactly one template. Remove sections that are not needed.

DIRECT

## Clarification Summary
### Mode
DIRECT

### Action
- ...

### Target
- ...

### Ready to Execute
- Yes / No
- Missing (if No): ...

### Safety Confirmation (only if risky/destructive)
- ...

DEBUG

## Clarification Summary
### Mode
DEBUG

### Problem
- Symptom: ...
- Expected vs Actual: ...
- Repro: ...

### Investigation
- Hypotheses:
  - ...
- Next Checks:
  1. ...
  2. ...

### Done Definition
- Root cause identified and fix validated.

IMPLEMENT

## Clarification Summary
### Mode
IMPLEMENT

### Clean Goal
- Outcome: ...
- User/Actor: ...

### Clean Spec
- Scope In: ...
- Out of Scope Now: ...
- Mandatory Requirements (testable):
  - ...
- Acceptance Criteria (testable):
  - AC1: Given ..., When ..., Then ...
- Constraints: ... (only if needed)

### Decisions Made
| Topic | Decision |
|------|----------|
| Scope | ... |
| Interface | ... |
| Constraints | ... |

PLAN

## Clarification Summary
### Mode
PLAN

### Goal
- ...

### Plan
1. ...
2. ...
3. ...

### Risks / Checkpoints (only if needed)
- ...

### Execution Decision
- Plan only by default. Ask before execution.

ASK

## Clarification Summary
### Mode
ASK

### Answer
- ...

### Evidence (optional)
- `path/to/file` or URL

RESEARCH

## Clarification Summary
### Mode
RESEARCH

### Findings
- ...

### Sources
- https://...

### Recommendation
- ...

OTHER

## Clarification Summary
### Mode
OTHER

### Hint
We have to create new format for this case so we can improve it.

### Next Step
- Define the new minimal output format for this case.

Guardrails

• Keep sections minimal and relevant.
• Ask only non-redundant blocker questions.
• Use constraints backed by user decisions or evidence.
• Preserve user intent.