Reactor — Strategic Codebase Re-direction

When the code works but the direction is wrong, you don't debug — you REACT. TRIZ-powered protocol for pivoting large codebases without losing stability.

When to Use

ALWAYS when:

• Requirements changed significantly after code was built
• Architecture no longer fits the problem (wrong patterns, wrong abstractions)
• 3+ patches on the same area — symptom of structural mismatch
• Tech debt blocks new feature development
• "We need to rewrite X" — STOP. Use this skill first.
• Migrating from one framework/library/pattern to another
• Post-mortem reveals systemic issues needing strategic change

Skip when:

• Simple bug fix → use cm-debugging
• Routine refactoring (extract method, rename) → use cm-clean-code
• New project from scratch → use cm-project-bootstrap
• Small scope change (< 5 files affected) → just refactor directly

The Iron Law

NO REWRITE WITHOUT REACTOR ANALYSIS FIRST

Rewrites fail 70% of the time. Incremental strategic migration succeeds 90% of the time.

TRIZ Principles Applied

#	Principle	How Applied
#35	Parameter Change	Change the fundamental parameter (language, pattern, architecture) — not the symptom
#28	Mechanics Substitution	Replace one mechanism with a more effective one (OOP → FP, REST → GraphQL, etc.)
#13	The Other Way Around	Instead of adapting new code to old architecture, adapt old architecture to new requirements
#25	Self-Service	Design the migration so each component can migrate independently
#1	Segmentation	Break monolithic change into independent migration units
#10	Prior Action	Prepare the codebase (interfaces, adapters) BEFORE the actual migration

The 5-Phase Process

Phase 1: ASSESS    → Understand what's wrong and why (not just symptoms)
Phase 2: MAP       → Trace all dependencies and blast radius
Phase 3: DESIGN    → Plan the migration path with strangler fig pattern
Phase 4: EXECUTE   → Migrate incrementally, one unit at a time
Phase 5: VERIFY    → Confirm direction is correct + clean up old code

Phase 1: ASSESS — Identify the Contradiction

Goal: Find the TRIZ contradiction — what do we WANT vs what BLOCKS us?

1. State the current direction:

   Current: We built [X architecture/pattern/structure]
   Problem: It doesn't support [Y requirement/scale/use case]
   

2. Identify the contradiction:

   We WANT: [desired capability]
   But: [current architecture] prevents it because [technical reason]
   
   TRIZ Contradiction:
   Improving [parameter A] worsens [parameter B]
   Example: "Improving modularity worsens performance"
           "Improving flexibility worsens type safety"
   

3. Define the Ideal Final Result (IFR):

   The system ITSELF [achieves the goal]
   WITHOUT [the current blocking factor]
   WHILE maintaining [what currently works well]
   

4. Scope assessment:

   Files affected: [count from codeintell or grep]
   Components affected: [list]
   Tests affected: [count]
   External API changes: [yes/no — breaking change?]
   Estimated effort: [S/M/L/XL]
   Risk level: [Low/Medium/High/Critical]
   

Phase 2: MAP — Dependency Analysis

Goal: Know exactly what touches what before changing anything.

1. Use Code Intelligence (if available):

   codegraph_impact("target_symbol", depth=3)
   → Shows all callers, dependencies, affected files
   
   codegraph_context("target_module")
   → Shows architecture around the area
   

2. Manual mapping (if codegraph unavailable):

   grep -rn "import.*{module}" src/
   grep -rn "{function_name}" src/
   → Build dependency tree manually
   

3. Categorize files by migration priority:

   Category	Description	Action
   Core	The files that MUST change for the new direction	Migrate first
   Dependent	Files that import/use Core files	Migrate after Core, use adapters
   Peripheral	Files loosely connected	Migrate last or leave untouched
   Dead	Files no longer needed after migration	Flag for deletion in Phase 5

4. Output: Migration Map Document

   ## Migration Map: [Initiative Name]
   
   ### Core (must change): [N files]
   - file_a.ts → [what changes]
   - file_b.ts → [what changes]
   
   ### Dependent (affected): [N files]
   - file_c.ts → [how affected]
   
   ### Peripheral (optional): [N files]
   - file_d.ts → [minimal change]
   
   ### Dead (remove after): [N files]
   - old_module.ts → DELETE after migration complete
   

Phase 3: DESIGN — Strangler Fig Migration

Goal: Design an incremental migration path — NEVER big-bang rewrite.

Strangler Fig Pattern (TRIZ #10 Prior Action):

1. Create NEW structure alongside OLD structure
2. Route NEW traffic/calls to NEW structure
3. Gradually migrate OLD consumers to NEW structure
4. Remove OLD structure when no longer used

Migration design template:

## Migration Path

### Step 1: Create Adapter Layer
- [ ] Create interface/abstraction that both old and new code satisfy
- [ ] All consumers now use the adapter, not direct implementation

### Step 2: Build New Implementation
- [ ] New code behind the adapter — can be tested independently
- [ ] Feature flag or config switch between old/new

### Step 3: Gradual Cut-over
- [ ] Migrate consumers one-by-one to new implementation
- [ ] Each migration is a separate commit/PR
- [ ] Tests pass at EVERY step (never break green)

### Step 4: Clean up (→ triggers cm-clean-code)
- [ ] Remove old implementation
- [ ] Remove adapter layer (if no longer needed)
- [ ] Remove feature flag
- [ ] Update documentation

Rules:

• Never break the build at any step
• Each step MUST be independently deployable
• Tests must pass at every intermediate state
• If any step fails → STOP, don't cascade

Phase 4: EXECUTE — Incremental Migration

Goal: Execute migration plan step by step with quality gates.

Use cm-execution Mode A (Batch) or Mode E (TRIZ-Parallel):

For each migration step:
  1. Write/update tests FIRST (cm-tdd)
  2. Implement the change
  3. Run full test suite
  4. Commit with clear message: "reactor: [step description]"
  5. If tests fail → STOP, diagnose, fix before next step
  
Progress tracking:
  → Update OpenSpec `tasks.md` and `cm-tasks.json` with each completed migration step
  → Update CONTINUITY.md with migration status

Commit convention:

reactor: create adapter layer for auth module
reactor: implement new auth service behind adapter
reactor: migrate login page to new auth
reactor: migrate signup page to new auth
reactor: remove old auth implementation
reactor: cleanup — remove adapter (direct usage now)

Phase 5: VERIFY & CLEAN

Goal: Confirm the new direction works AND trigger cm-clean-code.

1. Direction verification:

   □ New architecture supports the originally blocked requirement
   □ Performance is equal or better
   □ All tests pass
   □ No regression in existing features
   □ Documentation updated
   

2. Trigger cm-clean-code:

   After reactor completes → ALWAYS run cm-clean-code
   Reason: Migration leaves dead code, unused imports, stale references
   

3. Record in CONTINUITY.md:

   Decision: Migrated [X] from [old pattern] to [new pattern]
   Rationale: [why — the TRIZ contradiction we resolved]
   Scope: module:[affected module]
   

---

Red Flags — STOP

Thought	Reality
"Let's just rewrite everything"	Big-bang rewrites fail 70%+ of the time
"It's faster to start from scratch"	You lose all edge-case handling and bug fixes
"We don't need tests during migration"	Migration without tests = guaranteed regression
"Let's change the direction AND add features"	One thing at a time. Direction first, features after
"This adapter layer is extra work"	Adapter saves you from big-bang. It pays for itself
"We can just search-and-replace"	Structural changes ≠ text changes

Integration

Skill	When
cm-brainstorm-idea	UPSTREAM: Identifies need for direction change
cm-codeintell	Phase 2: Dependency analysis via codegraph
cm-planning	Phase 3: Formalize migration plan
cm-execution	Phase 4: Execute migration steps
cm-tdd	Phase 4: Tests before each migration step
cm-clean-code	Phase 5: MANDATORY cleanup after reactor
cm-debugging	If migration introduces bugs
cm-continuity	Record direction decisions

Lifecycle Position

cm-brainstorm-idea → cm-reactor → cm-planning → cm-execution → cm-clean-code
     (analyze)       (redirect)     (plan)        (build)        (hygiene)

The Bottom Line

Don't rewrite. React. Migrate incrementally. Clean up after. Every step must pass tests.