marsai:dev-refactor

Dev Refactor Skill

Analyzes existing codebase against MarsAI/V4-Company standards and generates refactoring tasks compatible with marsai:dev-cycle.

---

⛔ MANDATORY GAP PRINCIPLE (NON-NEGOTIABLE)

any divergence from MarsAI standards = MANDATORY gap to implement.

<cannot_skip>

• All divergences are gaps - Every difference MUST be tracked as FINDING-XXX
• Severity affects PRIORITY, not TRACKING - Low severity = lower priority, not "optional"
• No filtering allowed - You CANNOT decide which divergences "matter"
• No alternative patterns accepted - Different approach = STILL A GAP
• No cosmetic exceptions - Naming, formatting, structure differences = GAPS </cannot_skip>

Non-negotiable, not open to interpretation—a HARD RULE.

Anti-Rationalization: Mandatory Gap Principle

See shared-patterns/shared-anti-rationalization.md for:

• Refactor Gap Tracking section (mandatory gap principle rationalizations)
• Gate Execution section (workflow skip rationalizations)
• TDD section (test-first rationalizations)
• Universal section (general anti-patterns)

Verification Rule

COUNT(non-✅ items in all Standards Coverage Tables) == COUNT(FINDING-XXX entries)

If counts don't match → SKILL FAILURE. Go back and add missing findings.

---

Severity Calibration

Severity	Criteria	Examples
CRITICAL	Security risk, data corruption, production blocker	Exposed credentials, SQL injection, missing auth
HIGH	Major standards violation, architecture issue	Missing hexagonal architecture, wrong library usage
MEDIUM	Code quality, maintainability issues	Incomplete observability, naming conventions
LOW	Best practices, documentation	Missing comments, minor refactoring

MANDATORY: Track all severities. Severity affects priority of execution, not whether to track.

---

⛔ Architecture Pattern Applicability

Not all architecture patterns apply to all services. Before flagging gaps, verify the pattern is applicable.

Service Type	Hexagonal/Clean Architecture	Directory Structure
CRUD API (with services, adapters)	✅ APPLY	✅ APPLY (V4-Company pattern)
Complex business logic	✅ APPLY	✅ APPLY
Multiple bounded contexts	✅ APPLY	✅ APPLY
Event-driven systems	✅ APPLY	✅ APPLY
Simple scripts/utilities	❌ not APPLICABLE	❌ not APPLICABLE
CLI tools	❌ not APPLICABLE	❌ not APPLICABLE
Workers/background jobs	❌ not APPLICABLE	❌ not APPLICABLE
Simple lambda/functions	❌ not APPLICABLE	❌ not APPLICABLE

Detection Criteria

CRUD API (Hexagonal/V4-Company Pattern APPLICABLE):

• Service exposes API endpoints (REST, gRPC, GraphQL)
• Contains business logic and models
• Has CRUD operations (Create, Read, Update, Delete)
• Uses repositories for data access
• → MUST follow Hexagonal Architecture and V4-Company directory pattern

Simple Service (Hexagonal/V4-Company not applicable):

• CLI tools and scripts
• Workers and background jobs
• Simple utility functions
• Lambda functions with single responsibility
• No business logic layer

Agent Instruction

When dispatching specialist agents, include:

⛔ ARCHITECTURE APPLICABILITY CHECK:
1. If service is an API with CRUD operations → APPLY Hexagonal/V4-Company standards
2. If service is CLI tool, script, or simple utility → Do not flag Hexagonal/V4-Company gaps

CRUD APIs MUST follow Hexagonal Architecture (ports/adapters) and V4-Company directory pattern.

---

⛔ MANDATORY: Initialize Todo List FIRST

Before any other action, create the todo list with all steps:

TodoWrite:
  todos:
    - content: "Validate PROJECT_RULES.md exists"
      status: "pending"
      activeForm: "Validating PROJECT_RULES.md exists"
    - content: "Detect project stack (TypeScript backend only)"
      status: "pending"
      activeForm: "Detecting project stack"
    - content: "Read PROJECT_RULES.md for context"
      status: "pending"
      activeForm: "Reading PROJECT_RULES.md"
    - content: "Generate codebase report via marsai:codebase-explorer"
      status: "pending"
      activeForm: "Generating codebase report"
    - content: "Dispatch specialist agents in parallel"
      status: "pending"
      activeForm: "Dispatching specialist agents"
    - content: "Save individual agent reports"
      status: "pending"
      activeForm: "Saving agent reports"
    - content: "Map agent findings to FINDING-XXX entries"
      status: "pending"
      activeForm: "Mapping agent findings"
    - content: "Generate findings.md"
      status: "pending"
      activeForm: "Generating findings.md"
    - content: "Map findings 1:1 to REFACTOR-XXX tasks"
      status: "pending"
      activeForm: "Mapping findings to tasks (1:1)"
    - content: "Generate tasks.md"
      status: "pending"
      activeForm: "Generating tasks.md"
    - content: "Generate visual change report"
      status: "pending"
      activeForm: "Generating visual change report"
    - content: "Get user approval"
      status: "pending"
      activeForm: "Getting user approval"
    - content: "Save all artifacts"
      status: "pending"
      activeForm: "Saving artifacts"
    - content: "Handoff to marsai:dev-cycle"
      status: "pending"
      activeForm: "Handing off to marsai:dev-cycle"

This is NON-NEGOTIABLE. Do not skip creating the todo list.

---

⛔ CRITICAL: Specialized Agents Perform All Tasks

See shared-patterns/shared-orchestrator-principle.md for full ORCHESTRATOR principle, role separation, forbidden/required actions, step-to-agent mapping, and anti-rationalization table.

Summary: You orchestrate. Agents execute. If using Bash/Grep/Read to analyze code → STOP. Dispatch agent.

---

Step 1: Validate PROJECT_RULES.md

TodoWrite: Mark "Validate PROJECT_RULES.md exists" as in_progress

<block_condition>

• docs/PROJECT_RULES.md does not exist </block_condition>

If condition is true, output blocker and TERMINATE. Otherwise continue to Step 1.

Check: Does docs/PROJECT_RULES.md exist?

• YES → Mark todo as completed, continue to Step 1
• no → Output blocker and TERMINATE:

## BLOCKED: PROJECT_RULES.md Not Found

Cannot proceed without project standards baseline.

**Required Action:** Create `docs/PROJECT_RULES.md` with:
- Architecture patterns
- Code conventions
- Testing requirements
- Technology stack decisions

Re-run after file exists.

---

Step 1: Detect Project Stack

TodoWrite: Mark "Detect project stack (TypeScript)" as in_progress

⛔ SCOPE: BACKEND CODE ONLY. This skill analyzes backend code exclusively. MUST use marsai:dev-refactor-frontend for frontend code (React, Next.js, Vue, Angular).

Check for backend manifest files:

File/Pattern	Stack	Agent
package.json + Express/Fastify/NestJS (no React/Next.js)	TypeScript Backend	marsai:backend-engineer-typescript

Detection Logic:

• package.json exists + Express/Fastify/NestJS in dependencies (NO React/Next.js) → Add TypeScript backend agent
• package.json exists + React/Next.js in dependencies → STOP: This is a frontend project. Use marsai:dev-refactor-frontend instead.

⛔ FORBIDDEN: Dispatching marsai:frontend-engineer, marsai:frontend-designer, marsai:ui-engineer, marsai:qa-analyst-frontend, or marsai:frontend-bff-engineer-typescript from this skill. These are frontend agents and belong to marsai:dev-refactor-frontend.

TodoWrite: Mark "Detect project stack (TypeScript)" as completed

---

Step 2: Read PROJECT_RULES.md

TodoWrite: Mark "Read PROJECT_RULES.md for context" as in_progress

Read tool: docs/PROJECT_RULES.md

Extract project-specific conventions for agent context.

TodoWrite: Mark "Read PROJECT_RULES.md for context" as completed

---

Step 3: Generate Codebase Report

TodoWrite: Mark "Generate codebase report via marsai:codebase-explorer" as in_progress

⛔ MANDATORY: Use Task Tool with marsai:codebase-explorer

<dispatch_required agent="marsai:codebase-explorer"> Generate a comprehensive codebase report describing WHAT EXISTS.

Include:

• Project structure and directory layout
• Architecture pattern (hexagonal, clean, etc.)
• Technology stack from manifests
• Code patterns: config, database, handlers, errors, telemetry, testing
• Key files inventory with file:line references
• Code snippets showing current implementation patterns </dispatch_required>

<output_required>

EXPLORATION SUMMARY

[Your summary here]

KEY FINDINGS

[Your findings here]

ARCHITECTURE INSIGHTS

[Your insights here]

RELEVANT FILES

[Your file inventory here]

RECOMMENDATIONS

[Your recommendations here] </output_required>

Do not complete without outputting full report in the format above.

Anti-Rationalization Table for Step 3

See shared-patterns/anti-rationalization-codebase-explorer.md for the marsai:codebase-explorer dispatch anti-rationalization table.

FORBIDDEN Actions for Step 3

- Bash(command="find ... -name '*.ts'") → SKILL FAILURE - Bash(command="ls -la ...") → SKILL FAILURE - Bash(command="tree ...") → SKILL FAILURE - Task(subagent_type="Explore", ...) → SKILL FAILURE - Task(subagent_type="general-purpose", ...) → SKILL FAILURE - Task(subagent_type="Plan", ...) → SKILL FAILURE

Any of these = IMMEDIATE SKILL FAILURE.

REQUIRED Action for Step 3

✅ Task(subagent_type="marsai:codebase-explorer", ...)

Timestamp format: {timestamp} = YYYY-MM-DDTHH:MM:SS (e.g., 2026-02-07T22:30:45). Generate once at start, reuse for all artifacts.

After Task completes, save with Write tool:

Write tool:
  file_path: "docs/marsai:dev-refactor/{timestamp}/codebase-report.md"
  content: [Task output]

TodoWrite: Mark "Generate codebase report via marsai:codebase-explorer" as completed

---

Step 4: Dispatch Specialist Agents

TodoWrite: Mark "Dispatch specialist agents in parallel" as in_progress

⛔ File Size Enforcement (MANDATORY)

See shared-patterns/file-size-enforcement.md for thresholds and agent instructions.

All analysis agents MUST flag files exceeding 300 lines as ISSUE-XXX (converted to FINDING-XXX during Step 4.1 mapping, like all other agent issues). Files > 500 lines are CRITICAL severity. Files > 1000 lines are CRITICAL with mandatory decomposition plan included in the finding. Each oversized file = one ISSUE-XXX (not grouped). Test files are included — large test files are equally hard to maintain.

⛔ HARD GATE: Verify codebase-report.md Exists

BEFORE dispatching any specialist agent, verify:

Check 1: Does docs/marsai:dev-refactor/{timestamp}/codebase-report.md exist?
  - YES → Continue to dispatch agents
  - no  → STOP. Go back to Step 3.

Check 2: Was codebase-report.md created by marsai:codebase-explorer?
  - YES → Continue
  - no (created by Bash output) → DELETE IT. Go back to Step 3. Use correct agent.

If you skipped Step 3 or used Bash instead of Task tool → You MUST go back and redo Step 3 correctly.

Dispatch all applicable agents in ONE message (parallel):

⛔ MANDATORY: Reference Standards Coverage Table

All agents MUST follow shared-patterns/standards-coverage-table.md which defines:

• all sections to check per agent (including DDD)
• Required output format (Standards Coverage Table)
• Anti-rationalization rules
• Completeness verification

Section indexes are pre-defined in shared-patterns. Agents MUST check all sections listed.

---

For TypeScript Backend projects:

<parallel_dispatch agents="marsai:backend-engineer-typescript, marsai:qa-analyst, marsai:devops-engineer, marsai:sre"> All four agents MUST be dispatched in parallel via Task tool. Input: codebase-report.md, PROJECT_RULES.md </parallel_dispatch>

Task tool 1:
  subagent_type: "marsai:backend-engineer-typescript"
  description: "TypeScript backend standards analysis"
  prompt: |
    **MODE: ANALYSIS only**

    ⛔ MANDATORY: Check all sections in typescript.md per shared-patterns/standards-coverage-table.md

    ⛔ FRAMEWORKS & LIBRARIES DETECTION (MANDATORY):
    1. Read package.json to extract all dependencies used in codebase
    2. Load typescript.md standards via WebFetch → extract all listed frameworks/libraries
    3. For each category in standards (Backend Framework, ORM, Validation, Testing, etc.):
       - Compare codebase dependency vs standards requirement
       - If codebase uses DIFFERENT library than standards → ISSUE-XXX
       - If codebase is MISSING required library → ISSUE-XXX
    4. any library not in standards that serves same purpose = ISSUE-XXX

    ⛔ FILE SIZE ENFORCEMENT (MANDATORY):
    - Any source file > 300 lines (including test files, excluding auto-generated: *.d.ts, *.gen.ts, *.generated.ts, */generated/*, */__generated__/*) MUST be flagged as ISSUE-XXX
    - 301-500 lines: severity HIGH
    - > 500 lines: severity CRITICAL
    - > 1000 lines: severity CRITICAL with explicit decomposition plan
    - Include current line count and proposed split strategy in each finding
    - See shared-patterns/file-size-enforcement.md for split patterns
    - Reference: typescript.md → File Organization (MANDATORY)

    ⛔ MULTI-TENANT ANALYSIS (MANDATORY):
    See [shared-patterns/multi-tenant-analysis.md](../shared-patterns/multi-tenant-analysis.md) for the full checklist.

    Input:
    - MarsAI Standards: Load via WebFetch (typescript.md)
    - Section Index: See shared-patterns/standards-coverage-table.md → "marsai:backend-engineer-typescript"
    - Codebase Report: docs/marsai:dev-refactor/{timestamp}/codebase-report.md
    - Project Rules: docs/PROJECT_RULES.md

    Output:
    1. Standards Coverage Table (per shared-patterns format)
    2. ISSUE-XXX for each ⚠️/❌ finding with: Pattern name, Severity, file:line, Current Code, Expected Code

Agent Dispatch Summary

Stack Detected	Agents to Dispatch
TypeScript Backend	Task 1 (TS Backend) + Task 2-4

⛔ MUST use marsai:dev-refactor-frontend for frontend/BFF projects. This skill does not dispatch frontend agents.

TodoWrite: Mark "Dispatch specialist agents in parallel" as completed

---

Step 4.5: Save Individual Agent Reports

TodoWrite: Mark "Save individual agent reports" as in_progress

⛔ MANDATORY: Each agent's output MUST be saved as an individual report file.

After all parallel agent tasks complete, save each agent's output to a separate file:

docs/marsai:dev-refactor/{timestamp}/reports/
├── marsai:backend-engineer-typescript-report.md (always)
├── marsai:qa-analyst-report.md                  (always)
├── marsai:devops-engineer-report.md             (always)
└── marsai:sre-report.md                         (always)

Report File Format

Use Write tool for each agent report:

# {Agent Name} Analysis Report

**Generated:** {timestamp}
**Agent:** {agent-name}
**Mode:** ANALYSIS only

## Standards Coverage Table

{Copy agent's Standards Coverage Table output here}

## Issues Found

{Copy all ISSUE-XXX entries from agent output}

## Summary

- **Total Issues:** {count}
- **Critical:** {count}
- **High:** {count}
- **Medium:** {count}
- **Low:** {count}

---
*Report generated by marsai:dev-refactor skill*

Agent Report Mapping

Agent Dispatched	Report File Name
marsai:backend-engineer-typescript	marsai:backend-engineer-typescript-report.md
marsai:qa-analyst	marsai:qa-analyst-report.md
marsai:devops-engineer	marsai:devops-engineer-report.md
marsai:sre	marsai:sre-report.md

Anti-Rationalization Table for Step 4.5

Rationalization	Why It's WRONG	Required Action
"I'll combine all reports into one file"	Individual reports enable targeted re-runs and tracking	Save each agent to SEPARATE file
"Agent output is already visible in chat"	Chat history is ephemeral; files are artifacts	MUST persist as files
"Only saving reports with issues"	Empty reports prove compliance was checked	Save all dispatched agent reports
"findings.md already captures everything"	findings.md is processed; reports are raw agent output	Save BOTH raw reports and findings.md

REQUIRED Action for Step 4.5

Write tool:
  file_path: "docs/marsai:dev-refactor/{timestamp}/reports/{agent-name}-report.md"
  content: [Agent Task output formatted per template above]

Repeat for each agent dispatched in Step 4.

TodoWrite: Mark "Save individual agent reports" as completed

---

Step 4.1: Agent Report → Findings Mapping (HARD GATE)

TodoWrite: Mark "Map agent findings to FINDING-XXX entries" as in_progress

⛔ MANDATORY: all agent-reported issues MUST become findings.

Agent Report	Action
Any difference between current code and MarsAI standard	→ Create FINDING-XXX
Any missing pattern from MarsAI standards	→ Create FINDING-XXX
Any deprecated pattern usage	→ Create FINDING-XXX
Any observability gap	→ Create FINDING-XXX

FORBIDDEN Actions for Step 4.1

❌ Ignoring agent-reported issues because they seem "minor"  → SKILL FAILURE
❌ Filtering out issues based on personal judgment           → SKILL FAILURE
❌ Summarizing multiple issues into one finding              → SKILL FAILURE
❌ Skipping issues without ISSUE-XXX format from agent       → SKILL FAILURE
❌ Creating findings only for "interesting" gaps             → SKILL FAILURE

REQUIRED Actions for Step 4.1

✅ Every line item from agent reports becomes a FINDING-XXX entry
✅ Preserve agent's severity assessment exactly as reported
✅ Include exact file:line references from agent report
✅ Every non-✅ item in Standards Coverage Table = one FINDING-XXX
✅ Count findings in Step 5 MUST equal total issues from all agent reports

---

Anti-Rationalization Table for Step 4.1

⛔ See also: "Anti-Rationalization: Mandatory Gap Principle" at top of this skill.

Rationalization	Why It's WRONG	Required Action
"Multiple similar issues can be one finding"	Distinct file:line = distinct finding. Merging loses traceability.	One issue = One FINDING-XXX
"Agent report didn't use ISSUE-XXX format"	Format varies; presence matters. Every gap = one finding.	Extract all gaps into findings
"I'll consolidate to reduce noise"	Consolidation = data loss. Noise is signal.	Preserve all individual issues
"Some findings are duplicates across agents"	Different agents = different perspectives. Keep both.	Create separate findings per agent
"Team has approved this deviation"	Team approval ≠ standards compliance. Document the gap.	Create FINDING-XXX, note team decision
"Fixing this would break existing code"	Breaking risk = implementation concern, not tracking concern.	Create FINDING-XXX, note risk in description

⛔ MANDATORY GAP RULE FOR STEP 4.1

Per the Mandatory Gap Principle (see top of skill): any divergence from MarsAI standards = FINDING-XXX.

This means:

• ✅ items in Standards Coverage Table = No finding needed
• ⚠️ items = MUST create FINDING-XXX (partial compliance is a gap)
• ❌ items = MUST create FINDING-XXX (non-compliance is a gap)
• Different pattern = MUST create FINDING-XXX (alternative is still a gap)

Verification: Use formula from "Mandatory Gap Principle → Verification Rule" section.

⛔ Gate Escape Detection (Anti-Duplication)

When mapping findings, identify which gate SHOULD have caught the issue:

Finding Category	Should Be Caught In	Flag
Missing edge case tests	Gate 3 (Testing)	🚨 GATE 3 ESCAPE
Test isolation issues	Gate 3 (Testing)	🚨 GATE 3 ESCAPE
Skipped/assertion-less tests	Gate 3 (Testing)	🚨 GATE 3 ESCAPE
Test naming convention	Gate 3 (Testing)	🚨 GATE 3 ESCAPE
Missing test coverage	Gate 3 (Testing)	🚨 GATE 3 ESCAPE
TDD RED phase missing	Gate 3 (Testing)	🚨 GATE 3 ESCAPE
Implementation pattern gaps	Gate 0 (Implementation)	Normal finding
Standards compliance gaps	Gate 0 (Implementation)	Normal finding
Observability gaps	Gate 2 (SRE)	🚨 GATE 2 ESCAPE
Docker/DevOps gaps	Gate 1 (DevOps)	🚨 GATE 1 ESCAPE

Gate Escape Output Format:

### FINDING-XXX: [Issue Title] 🚨 GATE 3 ESCAPE

**Escaped From:** Gate 3 (Testing)
**Why It Escaped:** [Quality Gate check that should have caught this]
**Prevention:** [Specific check to add to Gate 3 exit criteria]

[Rest of finding format...]

Purpose: Track which issues escape which gates. If many GATE 3 ESCAPE findings occur, the Quality Gate checks need strengthening.

---

Summary Table (MANDATORY at end of findings.md):

## Gate Escape Summary

| Gate | Escaped Issues | Most Common Type |
|------|----------------|------------------|
| Gate 0 (Implementation) | N | [type] |
| Gate 1 (DevOps) | N | [type] |
| Gate 2 (SRE) | N | [type] |
| Gate 3 (Testing) | N | [type] |

**Action Required:** If any gate has >2 escapes, review that gate's exit criteria.

TodoWrite: Mark "Map agent findings to FINDING-XXX entries" as completed

---

Step 5: Generate findings.md

TodoWrite: Mark "Generate findings.md" as in_progress

⛔ HARD GATE: Verify All Issues Are Mapped

BEFORE creating findings.md, apply the Verification Rule from "Mandatory Gap Principle" section.

If counts don't match → STOP. Go back to Step 4.1. Map missing issues.

FORBIDDEN Actions for Step 5

❌ Creating findings.md with fewer entries than agent issues  → SKILL FAILURE
❌ Omitting file:line references from findings                → SKILL FAILURE
❌ Using vague descriptions instead of specific code excerpts → SKILL FAILURE
❌ Skipping "Why This Matters" section for any finding        → SKILL FAILURE
❌ Generating findings.md without reading all agent reports   → SKILL FAILURE

REQUIRED Actions for Step 5

✅ Every FINDING-XXX includes: Severity, Category, Agent, Standard reference
✅ Every FINDING-XXX includes: Current Code with exact file:line
✅ Every FINDING-XXX includes: MarsAI Standard Reference with URL
✅ Every FINDING-XXX includes: Required Changes as numbered actions
✅ Every FINDING-XXX includes: Why This Matters with Problem/Standard/Impact
✅ Total finding count MUST match total issues from Step 4.1

Anti-Rationalization Table for Step 5

Rationalization	Why It's WRONG	Required Action
"I'll add details later during implementation"	findings.md is the source of truth. Incomplete = useless.	Complete all sections for every finding
"Code snippet is too long to include"	Truncate to relevant lines, but never omit. Context is required.	Include code with file:line reference
"Standard URL is obvious, skip it"	Agents and humans need direct links. Nothing is obvious.	Include full URL for every standard
"Why This Matters is redundant"	It explains business impact. Standards alone don't convey urgency.	Write Problem/Standard/Impact for all
"Some findings are self-explanatory"	Self-explanatory to you ≠ clear to implementer.	Complete all sections without exception
"I'll group small findings together"	Each finding = one task in Step 6. findings.md = atomic issues.	One finding = one FINDING-XXX entry

Use Write tool to create findings.md:

⛔ CRITICAL: Every issue reported by agents in Step 4 MUST appear here as a FINDING-XXX entry.

# Findings: {project-name}

**Generated:** {timestamp}
**Total Findings:** {count}

## ⛔ Mandatory Gap Principle Applied

**all divergences from MarsAI standards are tracked below. No filtering applied.**

| Metric | Count |
|--------|-------|
| Total non-✅ items from agent reports | {X} |
| Total FINDING-XXX entries below | {X} |
| **Counts match?** | ✅ YES (REQUIRED) |

**Severity does not affect tracking - all gaps are mandatory:**
| Severity | Count | Priority | Tracking |
|----------|-------|----------|----------|
| Critical | {N} | Execute first | **MANDATORY** |
| High | {N} | Execute in current sprint | **MANDATORY** |
| Medium | {N} | Execute in next sprint | **MANDATORY** |
| Low | {N} | Execute when capacity | **MANDATORY** |

---

## FINDING-001: {Pattern Name}

**Severity:** Critical | High | Medium | Low (all MANDATORY)
**Category:** {architecture | testing | devops}
**Agent:** {agent-name}
**Standard:** {file}.md:{section}

### Current Code
```{lang}
// file: {path}:{lines}
{actual code}

MarsAI Standard Reference

Standard: {standards-file}.md → Section: {section-name} Pattern: {pattern-name} URL: https://raw.githubusercontent.com/V4-Company/marsai/main/dev-team/docs/standards/{file}.md

Required Changes

1. {action item 1 - what to change}
2. {action item 2 - what to add/remove}
3. {action item 3 - pattern to follow}

Why This Matters

• Problem: {what is wrong with current code}
• Standard Violated: {specific section from MarsAI standards}
• Impact: {business/technical impact if not fixed}

---

FINDING-002: ...

**TodoWrite:** Mark "Generate findings.md" as `completed`

---

## Step 6: Map Findings to Tasks (1:1)

**TodoWrite:** Mark "Map findings 1:1 to REFACTOR-XXX tasks" as `in_progress`

**⛔ HARD GATE: One FINDING-XXX = One REFACTOR-XXX task. No grouping.**

Each finding becomes its own task. This prevents findings from being lost inside grouped tasks.

**1:1 Mapping Rule:**
- FINDING-001 → REFACTOR-001
- FINDING-002 → REFACTOR-002
- FINDING-NNN → REFACTOR-NNN

**Ordemarsai:** Sort tasks by severity (Critical first), then by dependency order.

**Mapping Verification:**

Before proceeding to Step 7, verify:

• Total FINDING-XXX in findings.md: X
• Total REFACTOR-XXX in tasks.md: X (MUST MATCH exactly)
• Orphan findings (not mapped): 0 (MUST BE ZERO)
• Grouped tasks (multiple findings): 0 (MUST BE ZERO)

**If counts don't match → STOP. Every finding MUST have its own task.**

### Anti-Rationalization Table for Step 6

| Rationalization | Why It's WRONG | Required Action |
|-----------------|----------------|-----------------|
| "These findings are in the same file, I'll group them" | Grouping hides findings. One fix may be done, others forgotten. | **One finding = One task. No exceptions.** |
| "Grouping reduces task count and is easier to manage" | Fewer tasks = less visibility. Each finding needs independent tracking. | **Create one REFACTOR-XXX per FINDING-XXX** |
| "These are related and should be fixed together" | Related ≠ same task. Dev-cycle can execute them sequentially. | **Separate tasks, use Dependencies field to link** |
| "Too many tasks will overwhelm the developer" | Missing fixes overwhelms production. Completeness > convenience. | **Create all tasks. Priority handles ordering.** |

**TodoWrite:** Mark "Map findings 1:1 to REFACTOR-XXX tasks" as `completed`

---

## Step 7: Generate tasks.md

**TodoWrite:** Mark "Generate tasks.md" as `in_progress`

**Use Write tool to create tasks.md:**

```markdown
# Refactoring Tasks: {project-name}

**Source:** findings.md
**Total Tasks:** {count}

## ⛔ Mandatory 1:1 Mapping Verification

**Every FINDING-XXX has exactly one REFACTOR-XXX. No grouping.**

| Metric | Count |
|--------|-------|
| Total FINDING-XXX in findings.md | {X} |
| Total REFACTOR-XXX in tasks.md | {X} |
| **Counts match exactly?** | ✅ YES (REQUIRED) |
| Grouped tasks (multiple findings) | 0 (REQUIRED) |

**Priority affects execution order, not whether to include:**
- Critical/High tasks: Execute first
- Medium tasks: Execute in current cycle
- Low tasks: Execute when capacity - STILL MANDATORY TO COMPLETE

---

## REFACTOR-001: {Finding Pattern Name}

**Finding:** FINDING-001
**Severity:** Critical | High | Medium | Low (all ARE MANDATORY)
**Category:** {architecture | testing | devops}
**Agent:** {agent-name}
**Effort:** {hours}h
**Dependencies:** {other REFACTOR-XXX tasks or none}

### Current Code
```{lang}
// file: {path}:{lines}
{actual code from FINDING-001}

MarsAI Standard Reference

Standard File	Section	URL
{file}.md	{section}	Link

Required Actions

1. {action 1 - specific change to make}
2. {action 2 - pattern to implement}

Acceptance Criteria

• Code follows {standard}.md → {section} pattern
• No {anti-pattern} usage remains
• Tests pass after refactoring

**TodoWrite:** Mark "Generate tasks.md" as `completed`

---

## Step 7.5: Visual Change Report

**TodoWrite:** Mark "Generate visual change report" as `in_progress`

**MANDATORY:** Invoke `Skill("marsai:visual-explainer")` to produce a self-contained HTML page showing all planned refactoring changes. This replaces reading raw findings.md / tasks.md markdown for approval decisions.

**Read the code-diff template first:** Read `default/skills/visual-explainer/templates/code-diff.html` to absorb the patterns before generating.

**Generate the HTML report with these sections:**

### 1. Summary Dashboard
- Total FINDING-XXX count with severity breakdown (Critical / High / Medium / Low)
- Total files affected (unique file paths from all findings)
- Horizontal severity breakdown bar

### 2. Per-Finding Diff Panels (one section per FINDING-XXX)
For each FINDING-XXX in findings.md:
- **Header:** Finding ID, severity badge, category, agent that reported it
- **Before panel:** Current Code block from findings.md (with file:line reference, syntax highlighted via Highlight.js)
- **After panel:** MarsAI Standard pattern from Required Changes section (syntax highlighted)
- **Collapsible "Why This Matters":** Problem / Standard Violated / Impact from findings.md

### 3. Task Mapping Table
Table showing: FINDING-XXX → REFACTOR-XXX → Severity → Category → Estimated Effort

**Output:** Save to `docs/marsai:dev-refactor/{timestamp}/change-report.html`

**Open in browser:**
```text
macOS: open docs/marsai:dev-refactor/{timestamp}/change-report.html
Linux: xdg-open docs/marsai:dev-refactor/{timestamp}/change-report.html

Tell the user the file path. The report opens before the approval question so the user can review changes visually.

See shared-patterns/anti-rationalization-visual-report.md for anti-rationalization table.

TodoWrite: Mark "Generate visual change report" as completed

---

Step 8: User Approval

TodoWrite: Mark "Get user approval" as in_progress

<user_decision> MUST wait for explicit user response before proceeding. Options: Approve all | Critical only | Cancel </user_decision>

AskUserQuestion:
  questions:
    - question: "Review refactoring plan. How to proceed?"
      header: "Approval"
      options:
        - label: "Approve all"
          description: "Proceed to marsai:dev-cycle execution"
        - label: "Critical only"
          description: "Execute only Critical/High tasks"
        - label: "Cancel"
          description: "Keep analysis, skip execution"

CANNOT proceed without explicit user selection.

TodoWrite: Mark "Get user approval" as completed

---

Step 9: Save Artifacts

TodoWrite: Mark "Save all artifacts" as in_progress

docs/marsai:dev-refactor/{timestamp}/
├── codebase-report.md  (Step 3)
├── reports/            (Step 4.5)
│   ├── marsai:backend-engineer-typescript-report.md
│   ├── marsai:qa-analyst-report.md
│   ├── marsai:devops-engineer-report.md
│   └── marsai:sre-report.md
├── findings.md         (Step 5)
├── tasks.md           (Step 7)
└── change-report.html (Step 7.5)

TodoWrite: Mark "Save all artifacts" as completed

---

Step 10: Handoff to marsai:dev-cycle

TodoWrite: Mark "Handoff to marsai:dev-cycle" as in_progress

If user approved, use Skill tool to invoke marsai:dev-cycle directly:

Skill tool:
  skill: "marsai:dev-cycle"

⛔ CRITICAL: Pass tasks file path in context:

After invoking the skill, provide:

• Tasks file: docs/marsai:dev-refactor/{timestamp}/tasks.md

Context for marsai:dev-cycle:
  tasks-file: "docs/marsai:dev-refactor/{timestamp}/tasks.md"

Where {timestamp} format is YYYY-MM-DDTHH:MM:SS (e.g., 2026-02-07T22:30:45). Use the same timestamp across all artifacts in a single run.

Anti-Rationalization: Skill Invocation

Rationalization	Why It's WRONG	Required Action
"SlashCommand is equivalent to Skill tool"	SlashCommand is a hint; Skill tool guarantees skill loading	Use Skill tool, not SlashCommand
"User can run /marsai:dev-cycle manually"	Manual run risks skill not being loaded	Invoke Skill tool directly
"marsai:dev-cycle will auto-discover tasks"	Explicit path ensures correct file is used	Pass explicit tasks path
"User approved, I can skip marsai:dev-cycle"	Approval = permission to proceed, not skip execution	Invoke Skill tool
"Tasks are saved, job is done"	Saved tasks without execution = incomplete workflow	Invoke Skill tool

⛔ HARD GATE: You CANNOT complete marsai:dev-refactor without invoking Skill tool: marsai:dev-cycle.

If user approved execution, you MUST:

1. Invoke Skill tool: marsai:dev-cycle
2. Pass tasks file path: docs/marsai:dev-refactor/{timestamp}/tasks.md
3. Wait for marsai:dev-cycle to complete all 10 gates

Skipping this step = SKILL FAILURE.

marsai:dev-cycle executes each REFACTOR-XXX task through the 8-gate process.

TodoWrite: Mark "Handoff to marsai:dev-cycle" as completed

---

Execution Report

Base metrics per shared-patterns/output-execution-report.md.

Metric	Value
Duration	Xm Ys
Iterations	N
Result	PASS/FAIL/PARTIAL

Refactor-Specific Metrics

Metric	Value
Agents Dispatched	N
Findings Generated	N
Tasks Created	N
Artifacts Location	docs/marsai:dev-refactor/{date}/

Output Schema

artifacts:
  - codebase-report.md (Step 3)
  - reports/{agent-name}-report.md (Step 4.5)
  - findings.md (Step 5)
  - tasks.md (Step 7)
  - change-report.html (Step 7.5)

traceability:
  MarsAI Standard → Agent Report → FINDING-XXX → REFACTOR-XXX → Implementation