reinforce

id: reinforce name: reinforce description: | Add new requirements to spec.md or fill gaps in documents. Primary: Add features (replaces /enhance). Secondary: Fill gaps after /reverse. Triggers: reinforce, add requirements, new feature user-invocable: true version: 3.0.0 triggers: ["reinforce", "add requirements", "new feature", "supplement document", "fill gaps"] requires: [] platform: all recommended_model: sonnet allowed-tools: [Read, Write, Glob, LS, AskQuestion]

Global Rules: Adheres to rules/archflow-rules.md. Req ID FR-{number} Rule: Always use max(existing number) + 1. NEVER reuse deleted numbers.

Model: Sonnet (document merging). Opus for complex reasoning.

Reinforce Workflow

Add new requirements or fill gaps in existing documents.

Enhancement Workflow (replaces deleted /enhance)

New Proposal → /reinforce (spec.md) → /arch → /check → /build

When to Use

• Add new requirements: new features to spec.md, then run /arch
• After /reverse: fill gaps marked with ❓
• New information: reflect later-acquired info
• Error correction: fix incorrect inferences

Tool Fallback

Document Structure

docs/{serviceName}/
  ├── spec.md    # ← Input/Output
  ├── arch.md    # ← Input/Output
  └── trace.md

Phase 0: Skill Entry

0-1. Document Input

{
  "title": "Start Document Reinforcement",
  "questions": [
    {
      "id": "has_docs",
      "prompt": "Do you have documents to reinforce?",
      "options": [
        {"id": "yes", "label": "Yes - I will provide file paths with @"},
        {"id": "no", "label": "No - I don't have documents"}
      ]
    }
  ]
}

• no → guide to /reverse
• yes → request spec.md, arch.md paths

0-2. Extract serviceName

From path: docs/alert/spec.md → serviceName = "alert"

Phase 1: Analyze Existing Documents

1-1. Load Documents

Read spec.md and arch.md.

1-2. Identify Unconfirmed Items

Extract items marked with:

• ❓ Needs confirmation — requires user to provide content (delete and rewrite)
• ❓ Unconfirmed — not yet verified by user
• ❓ Inferred — AI-inferred, keep content but needs user confirmation to become ✅ Confirmed
• Items with "low" confidence
• Empty sections

NOTE: These three ❓ variants have different handling logic in Phase 4-3.

1-3. Report Current Status

## Current Document Status

### spec.md
| Section | Status |
|---------|--------|
| Goal | ✅ Confirmed / ❓ Inferred / ❌ Unconfirmed |
| Non-goals | {status} |
| Feature Specification | {status} |
| Priority | {status} |

### arch.md
| Section | Status |
|---------|--------|
| Summary | {status} |
| Code Mapping | ✅ Extracted |
| API Spec | ✅ Extracted |
| Risks & Tradeoffs | {status} |

### Items Needing Reinforcement
1. {Item 1} - {current status}
2. {Item 2} - {current status}

Phase 2: Collect New Information

2-1. Select Reinforcement Type

{
  "title": "What information do you want to add?",
  "questions": [
    {
      "id": "reinforce_type",
      "prompt": "Select the type of content to reinforce",
      "options": [
        {"id": "add_requirement", "label": "Add new requirements - New feature to spec.md (then run arch)"},
        {"id": "fill_blank", "label": "Fill gaps - Items marked with ❓"},
        {"id": "correct", "label": "Fix errors - Correct incorrect inferences"},
        {"id": "add_new", "label": "Add new info - Later-acquired information"},
        {"id": "guided", "label": "Guided mode - Ask me questions one by one"}
      ]
    }
  ]
}

2-2. Process by Type

add_requirement (PRIMARY USE):

1. Ask user for new requirement description
2. Read current Requirement Summary and find max Req ID:

   | Req ID | Category | Requirement | Priority | Status |
   |--------|----------|-------------|----------|--------|
   | FR-001 | Auth | ... | High | Designed |
   | FR-002 | Auth | ... | Medium | Implemented |
   

3. Assign new ID = max + 1 (e.g., FR-003). NEVER reuse deleted numbers.
4. Add row: | FR-003 | {category} | {new requirement} | {priority} | Draft |
5. Update related sections (Feature Specification, Non-Functional, etc.)
6. Inform: "Requirement FR-003 added. Next: run /arch to update design."

fill_blank: For each ❓ item, ask user (provide / skip). correct: User specifies section + correct content. add_new: User describes freely, classify to appropriate section. guided: Ask about unconfirmed items in sequence.

Phase 3: Classify Information

3-1. Analyze User Input

3-2. Confirm Classification (If Ambiguous)

{
  "title": "Classify Information",
  "questions": [
    {
      "id": "classify",
      "prompt": "Your input:\n\"{user input summary}\"\n\nWhere should this be reflected?",
      "options": [
        {"id": "requirements", "label": "Requirements (spec.md)"},
        {"id": "arch", "label": "Architecture (arch.md)"},
        {"id": "both", "label": "Both"},
        {"id": "auto", "label": "You decide"}
      ]
    }
  ]
}

Phase 4: Update Documents

4-1. Update spec.md

4-2. Update arch.md

Code Mapping Impl column rules:

• Existing rows: keep Impl status unchanged (do not modify [x] or [ ])
• New rows: set Impl = [ ] and continue # numbering
• Content modification: keep Impl as-is

4-3. Update Status Markers

Three ❓ variants have different transformation rules:

1. ❓ Needs confirmation → Content replacement (delete existing content, rewrite with user-provided info)
2. ❓ Inferred → ✅ Confirmed (keep content as-is, change status marker only)
3. Confidence low → high (metadata change only, if user confirmed)

4-4. Add Reinforcement Record

## Reinforcement History
| Date | Type | Changes |
|------|------|---------|
| {date} | fill_blank | Confirmed Goal, Non-goals |
| {date} | add_new | Added Risks section |

Phase 5: Save and Complete

5-1. Confirm Changes

## Changes to be Applied
### spec.md
| Section | Before | After |
|---------|--------|-------|
| {section} | {previous} | {new} |

### arch.md
| Section | Before | After |
|---------|--------|-------|
| {section} | {previous} | {new} |

Ask user: "I will update the documents with these changes. Proceed?"

5-2. Save

docs/{serviceName}/spec.md and docs/{serviceName}/arch.md

5-3. Completion Report

Report: changed sections, change type, completion rate before/after, remaining unconfirmed items. Next: need more → re-run /reinforce; complete → use sync/build.

Integration Flow

[reverse] → Incomplete documents
               ↓
         [reinforce] ←──┐
               │        │ (repeat)
               ↓        │
         Reinforced ────┘
         documents
               ↓ (when complete)
         Use sync/build

Important Notes

1. Works without /reverse — also for manual documents (but "fill gaps" unavailable without ❓ markers)
2. Incremental — re-run when new info available, no need to fill all at once
3. No code analysis — only processes docs + user input (token-efficient)
4. History logged — all reinforcements tracked in Reinforcement History section