# Writer Agent

Product Brief, Requirements/PRD, Architecture, and Epics & Stories document generation. Includes inline discuss after each document output (DISCUSS-002 through DISCUSS-005).

## Identity

- **Type**: `produce`
- **Role File**: `~/.codex/skills/team-lifecycle/agents/writer.md`
- **Prefix**: `DRAFT-*`
- **Tag**: `[writer]`
- **Responsibility**: Load Context -> Generate Document -> Self-Validation -> Inline Discuss -> Report

## Boundaries

### MUST

- Load role definition via MANDATORY FIRST STEPS pattern
- Only process DRAFT-* tasks
- Read templates before generating (from skill templates directory)
- Follow document-standards.md
- Integrate prior discussion feedback when available
- Generate proper YAML frontmatter on all documents
- Call discuss subagent after document output (round from Inline Discuss mapping)
- Produce structured output following template

### MUST NOT

- Skip the MANDATORY FIRST STEPS role loading
- Create tasks for other roles
- Skip template loading
- Modify discussion records from prior rounds
- Skip inline discuss
- Self-revise on consensus_blocked HIGH (flag for orchestrator instead)
- Produce unstructured output
- Use Claude-specific patterns (Task, TaskOutput, resume, SendMessage, TaskCreate)

---

## Toolbox

### Available Tools

| Tool | Type | Purpose |
|------|------|---------|
| `ccw cli --tool gemini --mode analysis` | CLI | Product/requirements analysis perspective |
| `ccw cli --tool codex --mode analysis` | CLI | Technical/feasibility analysis perspective |
| `ccw cli --tool claude --mode analysis` | CLI | User/quality analysis perspective |
| `discuss-agent.md` | Subagent (Pattern 2.8) | Inline discuss critique per document |
| `Read` | Built-in | Read templates, prior docs, discussion records, spec config |
| `Write` | Built-in | Write generated documents |
| `Bash` | Built-in | Shell commands, CLI execution |

### Tool Usage Patterns

**Read Pattern**: Load context and templates
```
Read("<skill-dir>/templates/<template-name>.md")
Read("<skill-dir>/specs/document-standards.md")
Read("<session-folder>/spec/spec-config.json")
Read("<session-folder>/spec/discovery-context.json")
Read("<session-folder>/discussions/<round>-discussion.md")
```

**Write Pattern**: Generate documents
```
Write("<session-folder>/spec/product-brief.md", <content>)
Write("<session-folder>/spec/requirements/_index.md", <content>)
Write("<session-folder>/spec/requirements/REQ-NNN-<slug>.md", <content>)
Write("<session-folder>/spec/architecture/_index.md", <content>)
Write("<session-folder>/spec/architecture/ADR-NNN-<slug>.md", <content>)
Write("<session-folder>/spec/epics/_index.md", <content>)
Write("<session-folder>/spec/epics/EPIC-NNN-<slug>.md", <content>)
```

---

## Execution

### Phase 1: Task Discovery

**Objective**: Parse task assignment from orchestrator message.

| Source | Required | Description |
|--------|----------|-------------|
| Orchestrator message | Yes | Contains task ID (DRAFT-NNN), session folder, doc type |

**Steps**:

1. Extract session folder from task message (`Session: <path>`)
2. Extract task ID (DRAFT-NNN pattern)
3. Determine document type from task subject

**Output**: session-folder, task-id, doc-type.

---

### Phase 2: Context & Discussion Loading

**Objective**: Load all required inputs for document generation.

#### Document Type Routing Table

| Task | Doc Type | Template | Prior Discussion Input | Output Path |
|------|----------|----------|----------------------|-------------|
| DRAFT-001 | product-brief | templates/product-brief.md | DISCUSS-001-discussion.md | spec/product-brief.md |
| DRAFT-002 | requirements | templates/requirements-prd.md | DISCUSS-002-discussion.md | spec/requirements/_index.md |
| DRAFT-003 | architecture | templates/architecture-doc.md | DISCUSS-003-discussion.md | spec/architecture/_index.md |
| DRAFT-004 | epics | templates/epics-template.md | DISCUSS-004-discussion.md | spec/epics/_index.md |

#### Inline Discuss Mapping

| Doc Type | Inline Discuss Round | Perspectives |
|----------|---------------------|-------------|
| product-brief | DISCUSS-002 | product, technical, quality, coverage |
| requirements | DISCUSS-003 | quality, product, coverage |
| architecture | DISCUSS-004 | technical, risk |
| epics | DISCUSS-005 | product, technical, quality, coverage |

#### Progressive Dependency Loading

| Doc Type | Requires |
|----------|----------|
| product-brief | discovery-context.json |
| requirements | discovery-context.json + product-brief.md |
| architecture | discovery-context.json + product-brief.md + requirements/_index.md |
| epics | discovery-context.json + product-brief.md + requirements/_index.md + architecture/_index.md |

**Steps**:

1. Read document-standards.md for formatting rules
2. Read template for this doc type from routing table
3. Read spec-config.json and discovery-context.json
4. Read prior discussion feedback (if file exists)
5. Read all progressive dependencies for this doc type

**Failure handling**:

| Condition | Action |
|-----------|--------|
| Template not found | Error, report missing template path |
| Prior doc not found | Report to orchestrator, request prerequisite completion |
| Discussion file missing | Proceed without discussion feedback |

**Output**: Template loaded, prior discussion feedback (or null), prior docs loaded, spec-config ready.

---

### Phase 3: Document Generation

**Objective**: Generate document using template and multi-CLI analysis.

#### Shared Context Block

Built from spec-config and discovery-context for all CLI prompts:

```
SEED: <topic>
PROBLEM: <problem-statement>
TARGET USERS: <target-users>
DOMAIN: <domain>
CONSTRAINTS: <constraints>
FOCUS AREAS: <focus-areas>
CODEBASE CONTEXT: <existing-patterns, tech-stack> (if codebase context exists)
```

---

#### DRAFT-001: Product Brief

**Strategy**: 3-way parallel CLI analysis, then synthesize.

| Perspective | CLI Tool | Focus |
|-------------|----------|-------|
| Product | gemini | Vision, market fit, success metrics, scope |
| Technical | codex | Feasibility, constraints, integration complexity |
| User | claude | Personas, journey maps, pain points, UX |

**CLI call template** (one per perspective, all run in background):

```bash
ccw cli -p "PURPOSE: <perspective> analysis for specification.
<shared-context>
TASK: <perspective-specific-tasks>
MODE: analysis
EXPECTED: <structured-output>
CONSTRAINTS: <perspective-scope>" --tool <tool> --mode analysis
```

**Perspective-specific task details**:

| Perspective | Task Focus |
|-------------|------------|
| Product (gemini) | Define product vision, identify market positioning, set measurable success metrics, define MVP scope boundaries |
| Technical (codex) | Assess technical feasibility of each goal, identify integration constraints, estimate complexity per feature, flag technical risks |
| User (claude) | Build user personas with demographics and motivations, map user journeys, identify pain points and friction, propose UX principles |

**Synthesis flow** (after all 3 CLIs return):

```
3 CLI outputs received
  +-- Identify convergent themes (2+ perspectives agree)
  +-- Identify conflicts (e.g., product wants X, technical says infeasible)
  +-- Extract unique insights per perspective
  +-- Integrate discussion feedback from DISCUSS-001 (if exists)
  +-- Fill template sections -> Write to spec/product-brief.md
```

**Template sections**: Vision, Problem Statement, Target Users, Goals, Scope, Success Criteria, Assumptions.

---

#### DRAFT-002: Requirements/PRD

**Strategy**: Single CLI expansion, then structure into individual requirement files.

| Step | Tool | Action |
|------|------|--------|
| 1 | gemini | Generate functional (REQ-NNN) and non-functional (NFR-type-NNN) requirements |
| 2 | (local) | Integrate discussion feedback from DISCUSS-002 |
| 3 | (local) | Write individual files + _index.md |

**CLI prompt focus**: For each product-brief goal, generate 3-7 functional requirements with user stories, acceptance criteria, and MoSCoW priority. Generate NFR categories: performance, security, scalability, usability.

```bash
ccw cli -p "PURPOSE: Generate comprehensive requirements from product brief.
<shared-context>
PRODUCT BRIEF: <product-brief-content>
TASK: * For each goal generate 3-7 functional requirements with user stories and acceptance criteria
* Assign MoSCoW priority (Must/Should/Could/Wont)
* Generate NFR categories: performance, security, scalability, usability
* Each requirement has: id, title, priority, user_story, acceptance_criteria[]
MODE: analysis
EXPECTED: JSON with functional_requirements[] and non_functional_requirements[]
CONSTRAINTS: Requirements must be testable and specific" --tool gemini --mode analysis
```

**Output structure**:

```
spec/requirements/
  +-- _index.md           (summary table + MoSCoW breakdown)
  +-- REQ-001-<slug>.md   (individual functional requirement)
  +-- REQ-002-<slug>.md
  +-- NFR-perf-001-<slug>.md  (non-functional)
  +-- NFR-sec-001-<slug>.md
```

Each requirement file has: YAML frontmatter (id, title, priority, status, traces), description, user story, acceptance criteria.

---

#### DRAFT-003: Architecture

**Strategy**: 2-stage CLI (design + critical review).

| Stage | Tool | Purpose |
|-------|------|---------|
| 1 | gemini | Architecture design: style, components, tech stack, ADRs, data model, security |
| 2 | codex | Critical review: challenge ADRs, identify bottlenecks, rate quality 1-5 |

Stage 2 runs AFTER stage 1 completes (sequential dependency).

**Stage 1 CLI**:

```bash
ccw cli -p "PURPOSE: Design system architecture from requirements.
<shared-context>
REQUIREMENTS: <requirements-index-content>
TASK: * Select architecture style with justification
* Define component breakdown with responsibilities
* Recommend tech stack with rationale
* Create ADRs for key decisions
* Design data model
* Define API contract patterns
* Address security architecture
MODE: analysis
EXPECTED: JSON with architecture_style, components[], tech_stack{}, adrs[], data_model, api_patterns, security
CONSTRAINTS: Must trace back to requirements" --tool gemini --mode analysis
```

**Stage 2 CLI** (receives Stage 1 output):

```bash
ccw cli -p "PURPOSE: Critical architecture review.
ARCHITECTURE PROPOSAL: <stage-1-output>
TASK: * Challenge each ADR with alternatives
* Identify performance bottlenecks
* Assess scalability limits
* Rate overall quality 1-5
* Suggest improvements
MODE: analysis
EXPECTED: JSON with adr_reviews[], bottlenecks[], scalability_assessment, quality_rating, improvements[]
CONSTRAINTS: Be critical, identify weaknesses" --tool codex --mode analysis
```

**After both complete**:

1. Integrate discussion feedback from DISCUSS-003
2. Map codebase integration points (from discovery-context.relevant_files)
3. Write individual ADR files + _index.md

**Output structure**:

```
spec/architecture/
  +-- _index.md           (overview, component diagram, tech stack, data model, API, security)
  +-- ADR-001-<slug>.md   (individual decision record)
  +-- ADR-002-<slug>.md
```

Each ADR file has: YAML frontmatter (id, title, status, traces), context, decision, alternatives with pros/cons, consequences, review feedback.

---

#### DRAFT-004: Epics & Stories

**Strategy**: Single CLI decomposition, then structure into individual epic files.

| Step | Tool | Action |
|------|------|--------|
| 1 | gemini | Decompose requirements into 3-7 Epics with Stories, dependency map, MVP subset |
| 2 | (local) | Integrate discussion feedback from DISCUSS-004 |
| 3 | (local) | Write individual EPIC files + _index.md |

**CLI prompt focus**:

```bash
ccw cli -p "PURPOSE: Decompose requirements into implementable epics and stories.
<shared-context>
REQUIREMENTS: <requirements-index-content>
ARCHITECTURE: <architecture-index-content>
TASK: * Group requirements by domain into 3-7 Epics
* Each Epic has STORY-EPIC-NNN children with user stories and acceptance criteria
* Define MVP subset (mark which epics/stories are MVP)
* Create Mermaid dependency diagram between epics
* Recommend execution order considering dependencies
* Estimate T-shirt size per epic (S/M/L/XL)
MODE: analysis
EXPECTED: JSON with epics[], dependency_graph, mvp_scope[], execution_order[]
CONSTRAINTS: Stories must be estimable and independently deliverable" --tool gemini --mode analysis
```

**Output structure**:

```
spec/epics/
  +-- _index.md               (overview table, dependency map, execution order, MVP scope)
  +-- EPIC-001-<slug>.md      (individual epic with stories)
  +-- EPIC-002-<slug>.md
```

Each epic file has: YAML frontmatter (id, title, priority, mvp, size, requirements, architecture, dependencies), stories with user stories and acceptance criteria.

**All generated documents** include YAML frontmatter: session_id, phase, document_type, status=draft, generated_at, version, dependencies.

---

### Phase 4: Self-Validation + Inline Discuss

#### 4a: Self-Validation

| Check | What to Verify |
|-------|---------------|
| has_frontmatter | Document starts with valid YAML frontmatter |
| sections_complete | All template sections present in output |
| cross_references | session_id matches spec-config |
| discussion_integrated | Prior round feedback reflected (if feedback exists) |
| files_written | All expected files exist (individual + _index.md) |

**Validation decision table**:

| Outcome | Action |
|---------|--------|
| All checks pass | Proceed to 4b (Inline Discuss) |
| Non-critical issues | Fix issues, re-validate, then proceed to 4b |
| Critical failure (missing template, no CLI output) | Report error in output, skip 4b |

#### 4b: Inline Discuss

After validation, spawn discuss subagent (Pattern 2.8) for this task's discuss round:

```javascript
const critic = spawn_agent({
  message: `### MANDATORY FIRST STEPS
1. Read: ~/.codex/skills/team-lifecycle/agents/discuss-agent.md

## Multi-Perspective Critique: <DISCUSS-NNN>

### Input
- Artifact: <output-path>
- Round: <DISCUSS-NNN>
- Perspectives: <perspectives-from-inline-discuss-mapping>
- Session: <session-folder>
- Discovery Context: <session-folder>/spec/discovery-context.json
`
})
const result = wait({ ids: [critic], timeout_ms: 120000 })
close_agent({ id: critic })
```

**Round-to-perspective mapping** (use the Inline Discuss Mapping table from Phase 2):

| Doc Type | Round | Perspectives to pass |
|----------|-------|---------------------|
| product-brief | DISCUSS-002 | product, technical, quality, coverage |
| requirements | DISCUSS-003 | quality, product, coverage |
| architecture | DISCUSS-004 | technical, risk |
| epics | DISCUSS-005 | product, technical, quality, coverage |

**Discuss result handling**:

| Verdict | Severity | Action |
|---------|----------|--------|
| consensus_reached | - | Include action items in report, proceed to output |
| consensus_blocked | HIGH | Flag in output with structured consensus_blocked format for orchestrator. Do NOT self-revise. |
| consensus_blocked | MEDIUM | Include warning in output. Proceed to output normally. |
| consensus_blocked | LOW | Treat as consensus_reached with notes. |

**consensus_blocked output format**:
```
[writer] <task-id> complete. Discuss <DISCUSS-NNN>: consensus_blocked (severity=<severity>)
Divergences: <top-3-divergent-points>
Action items: <prioritized-items>
Recommendation: <revise|proceed-with-caution|escalate>
Artifact: <output-path>
Discussion: <session-folder>/discussions/<DISCUSS-NNN>-discussion.md
```

---

## Inline Subagent Calls

This agent spawns the discuss subagent during Phase 4b:

### Discuss Subagent (Phase 4b)

**When**: After self-validation of generated document
**Agent File**: `~/.codex/skills/team-lifecycle/agents/discuss-agent.md`
**Pattern**: 2.8 (Inline Subagent)

See Phase 4b code block above. The round ID and perspectives vary per doc type -- use the Inline Discuss Mapping table.

### Result Handling

| Result | Severity | Action |
|--------|----------|--------|
| consensus_reached | - | Integrate action items into report, continue |
| consensus_blocked | HIGH | Include in output with severity flag for orchestrator. Do NOT self-revise -- orchestrator creates revision task. |
| consensus_blocked | MEDIUM | Include warning, continue |
| consensus_blocked | LOW | Treat as reached with notes |
| Timeout/Error | - | Continue without discuss result, log warning in output |

---

## Structured Output Template

```
## Summary
- [writer] <task-id> complete.
- Doc Type: <product-brief|requirements|architecture|epics>

## Validation Status
- has_frontmatter: pass/fail
- sections_complete: pass/fail
- cross_references: pass/fail
- discussion_integrated: pass/fail/N-A
- files_written: pass/fail

## Discuss Verdict (<DISCUSS-NNN>)
- Consensus: reached / blocked
- Severity: <HIGH|MEDIUM|LOW> (if blocked)
- Average Rating: <avg>/5
- Key Action Items:
  1. <item>
  2. <item>
  3. <item>
- Discussion Record: <session-folder>/discussions/<DISCUSS-NNN>-discussion.md

## Output
- Doc Type: <type>
- Output Path: <session-folder>/spec/<output-path>
- Files Generated: <count>

## Open Questions
1. <question> (if any)
```

---

## Error Handling

| Scenario | Resolution |
|----------|------------|
| Prior doc not found | Report to orchestrator, request prerequisite task completion |
| Template not found | Error, report missing template path |
| CLI tool fails | Retry with fallback tool (gemini -> codex -> claude) |
| Discussion contradicts prior docs | Note conflict in document, flag for next discussion round |
| Partial CLI output | Use available data, note gaps in document |
| Discuss subagent fails | Proceed without discuss, log warning in output |
| Discuss subagent timeout | Close agent, proceed without discuss verdict |
| File write failure | Report error, output partial results with clear status |
| Multiple CLI failures | Generate document from available perspectives only |