Se 1 Prompt Analysis

# SE Phase 1: Prompt Analysis

You are executing **SE Pipeline Phase 1: Prompt Analysis** for the feature described by the user.

## Phase Purpose

<!-- PIPELINE-STATE-2026-0001/0002/0003: write Step C deliverable to .claude/pipeline-state/<run-dir>/phase-<N>-<slug>.md; update manifest at Step D; read prior phase from disk at Step A. See specs/pipeline-state-persistence.md and .claude/pipeline-state/SCHEMA.md. -->

Deconstruct the user's prompt to identify all possible interpretations, scope boundaries, ambiguities, and implicit assumptions before any engineering work begins. This prevents downstream rework caused by misunderstood intent.

## Progress Reporting (MANDATORY)

Before each sub-step, output a progress line to the terminal:

─────────────────────────────────────────────────────── SE Phase 1: Prompt Analysis | Step X: [Step Name] ───────────────────────────────────────────────────────

Use these step names:
- `Step A: Discussion & Ideation (Tri-Persona)`
- `Step B: Critical Thinking Convergence`
- `Step C: Deliverable Generation`
- `Step D: Phase Approval`

At step completion, output: `  ✓ Phase 1 Step X complete → Proceeding to Step Y`
On rejection: `  ✗ Phase 1 Step D: REJECTED → Restarting from Step A (FREE restart)`

---

## Sub-Step Execution Protocol

Execute all 4 sub-steps sequentially. Each sub-step spawns a subagent via the **Task tool** with `subagent_type: "general-purpose", model: "opus"`.

---

### Step A: Discussion & Ideation (Tri-Persona Protocol)

Execute the **Tri-Persona Discussion protocol** (defined in `/se-step-a-discussion`) with:

- `$PHASE_NUMBER` = 1
- `$PHASE_NAME` = "Prompt Analysis"
- `$PHASE_CONTEXT` = Phase 0 Codebase Context Report (`$PHASE_0_DELIVERABLE`) + The original user feature prompt (`$FEATURE`) + `$ACCUMULATED_FEEDBACK`
- `$DISCUSSION_TOPIC` = "Analyze the user's prompt: What are the possible interpretations, hidden assumptions, scope boundaries, and risks? What is the user REALLY asking for?"

The three personas (Innovator, Guardian, Catalyst) will deliberate for a minimum of 2 rounds, continuing until all three declare CONVERGED or 5 rounds are reached.

### After Step A Returns

Display the full Tri-Persona Discussion Summary to the user. The Catalyst's Recommended Direction becomes the primary input for Step B. Proceed to Step B.

---

### Step B: Critical Thinking Convergence

Spawn a subagent with the following prompt (include Step A output):

---

**Persona:** You are the **Semantic Convergence Analyst**. You take a brainstormed analysis and converge it into a single, definitive interpretation. You are decisive and eliminate ambiguity ruthlessly.

**Step A Output:** [Include full Step A output]

**Your Task:**

1. **Resolve Intent** — Select the single most likely interpretation. Justify why alternatives are less likely.
2. **Classify Scope** — Produce a definitive IN/OUT scope list. Move all ambiguous items to either IN or OUT with reasoning.
3. **Flag Assumptions** — Mark each assumption as CONFIRMED (from codebase evidence), LIKELY (reasonable inference), or NEEDS_CLARIFICATION.
4. **Synthesize Requirements** — Produce a clean, numbered list of requirements derived from the resolved interpretation.

**Output Format:**

【Phase 1B: Semantic Convergence Report】

Resolved Intent

[1-2 sentence definitive statement of what the user wants]

Interpretation Selection

• Selected: [Interpretation X]
• Rejected: [Interpretation Y — reason], [Interpretation Z — reason]

Definitive Scope

IN Scope	OUT Scope
[Items]	[Items]

Assumption Status

#	Assumption	Status	Evidence
1	[Assumption]	CONFIRMED/LIKELY/NEEDS_CLARIFICATION	[Evidence]

Synthesized Requirements

1. [REQ-1] [Description]
2. [REQ-2] [Description] ...

Remaining Clarifications Needed

1. [Question — if any, max 3]

---

### After Step B Returns

Display the Convergence Report to the user.

**Assumption Checkpoint (MANDATORY):** Execute the **Assumption Checkpoint Protocol** (defined in se-pipeline.md). Extract all assumptions from the Convergence Report's "Assumption Status" table — specifically items marked LIKELY or NEEDS_CLARIFICATION. For each, extract the 4 fields (Assumption, Confidence, Current basis, Risk if wrong) and present the 5-column checkpoint table. Yield with "Awaiting your confirmation or amendments before proceeding." and wait for the user's response. Record decisions as Validated Assumptions (6 fields per item).

If zero assumptions require validation: output "No assumptions identified — proceeding to Step C." — no table, no yield.

Pass the validated assumptions as explicit input to Step C alongside the Step A and Step B outputs. These validated assumptions will propagate through the Phase 1 deliverable into Phase 3 and all subsequent phases.

Proceed to Step C.

---

### Step C: Deliverable Generation

Spawn a subagent with the following prompt (include Step A + B outputs + user checkpoint decisions):

---

**Persona:** You are the **Analysis Document Architect**. You produce the formal Phase 1 deliverable: a structured Prompt Analysis Document that downstream phases will consume.

**Step A + B Outputs + User Checkpoint Decisions:** [Include all]

**Your Task:** Produce the **Prompt Analysis Document** — the formal deliverable of Phase 1.

**Output Format:**

【Phase 1 Deliverable: Prompt Analysis Document】

1. Feature Summary

Title: [Short descriptive title] Intent: [1-2 sentence resolved intent from Step B] Scope Classification: [SMALL / MEDIUM / LARGE / CROSS-CUTTING]

2. Requirements Inventory

ID	Requirement	Priority	Source
REQ-1	[Description]	P1/P2/P3	[Prompt/Inferred/Codebase]

3. Scope Contract

Boundary	Decision	Rationale
[Item]	IN/OUT	[Why]

4. Assumptions Register

| ```