| name | explore-modularization-improvement-opportunities |
| description | Explore a codebase to find opportunities for architectural improvement, focusing on making the codebase more testable by deepening shallow modules. Use when user wants to improve architecture, find refactoring opportunities, consolidate tightly-coupled modules, or make a codebase more AI-navigable. |
Explore Modularization Improvement Opportunities
Explore a codebase like an AI would, surface architectural friction, discover opportunities for improving testability, and propose module-deepening refactors as GitHub issue RFCs.
A deep module (John Ousterhout, "A Philosophy of Software Design") has a small interface hiding a large implementation. Deep modules are more testable, more AI-navigable, and let you test at the boundary instead of inside. Assess depth primarily at the caller-facing boundary, not by counting internal files or private helpers.
Process
1. Explore the codebase
Use the Agent tool with subagent_type=Explore or the @explore agent to navigate the codebase naturally. Do NOT follow rigid heuristics. Explore organically, note where you experience friction, and treat that friction as a clue to investigate rather than proof that a module is shallow:
- Where does understanding one concept require bouncing between many small files?
- Where do public interfaces expose nearly as much complexity as they hide?
- Where have pure functions been extracted just for testability, but the real bugs hide in how they're called?
- Where do public boundaries between modules create integration risk for callers?
- Which parts of the codebase are untested, or hard to test?
Friction during exploration can reveal candidate areas, but do not present a module as shallow until you have verified that the caller-facing interface is wide, leaky, or forces callers to coordinate too much detail themselves.
Assess module depth
Before calling a module shallow or proposing a deepening refactor, inspect the boundary that external callers must understand.
Count as evidence of shallowness:
- Public types, functions, methods, commands, or schemas that external callers must learn
- Integration points and call patterns that leak internal coordination to callers
- Trivial wrappers that expose many entry points without hiding meaningful complexity
Do not count as evidence of shallowness by itself:
- Internal file count or private submodule structure
- Private helpers, internal data structures, or implementation-only indirection
- Language-specific internal visibility constructs that are not part of the real external contract
- Parameter count alone (many independent configuration values being passed through does not constitute coordination burden)
Heuristics:
- A deep module usually has a narrow boundary that hides substantial behavior or decision-making
- A shallow module often makes callers assemble workflows, pass many coordinated values, or understand multiple low-level concepts
- A small public API can still be shallow if callers must manage the real complexity themselves
- A fragmented internal implementation can be a maintainability smell, but it is not automatically a module-depth problem
2. Present candidates
Present a numbered list of deepening opportunities. Limit to 3-5 highest-impact candidates — if you find more, filter by: integration risk × test coverage gap × frequency of changes to the affected modules.
For each candidate, show:
- Cluster: Which modules/concepts are involved
- Why they're coupled: Shared types, call patterns, co-ownership of a concept
- Boundary evidence: Which caller-facing API surface or usage patterns make this candidate shallow, leaky, or coordination-heavy
- Dependency category: See DEPENDENCIES.md for the four categories
- Test impact: What existing tests would be replaced by boundary tests
Before presenting a candidate, verify:
- The concern is visible at a caller-facing boundary, not just inside the implementation
- The candidate exposes meaningful complexity to callers rather than hiding it well already
- The implementation contains non-trivial behavior worth hiding behind a smaller or clearer interface
- You are not flagging the module solely because it spans many files or feels annoying to read
Do NOT propose interfaces yet. Ask the user: "Which of these would you like to explore?"
3. User picks a candidate
4. Frame the problem space
Write a user-facing explanation of the problem space for the chosen candidate:
- The constraints any new interface would need to satisfy
- The dependencies it would need to rely on
- A rough illustrative code sketch to make the constraints concrete — this is not a proposal, just a way to ground the constraints
Ask for the user for feedback or confirmation.
WAIT for explicit user confirmation before proceeding to the next step.
5. Design multiple interfaces
Spawn 3+ sub-agents in parallel using the Agent tool. Each must produce a radically different interface for the deepened module.
Prompt each sub-agent with a separate technical brief (file paths, coupling details, dependency category, what's being hidden). This brief is independent of the user-facing explanation in Step 4. Give each agent a different design constraint:
- Agent 1: "Minimize the interface — aim for 1-3 entry points max"
- Agent 2: "Maximize flexibility — support many use cases and extension"
- Agent 3: "Optimize for the most common caller — make the default case trivial"
- Agent 4 (if applicable): "Design around the ports & adapters pattern for cross-boundary dependencies"
Each sub-agent outputs:
- Interface signature (types, methods, params)
- Usage example showing how callers use it
- What complexity it hides internally
- Dependency strategy (how deps are handled — see DEPENDENCIES)
- Trade-offs
Present designs sequentially, then compare them in prose.
Decision criteria — weigh designs by (in order):
- Testability improvement: Does it reduce the number of test scenarios needed while maintaining coverage?
- Caller simplicity: Does it reduce call-sites and cognitive load at the boundary?
- Migration feasibility: Can existing callers migrate incrementally without breaking changes?
- Flexibility: Is the interface extensible without modification (Open/Closed)?
After comparing, give your own recommendation: which design you think is strongest and why. If elements from different designs would combine well, propose a hybrid. Be opinionated — the user wants a strong read, not just a menu.
Output format: Each design must explicitly map to TASK_TEMPLATE.md sections:
- Interface signature → "Proposed Interface"
- Dependency strategy → "Dependency Strategy"
- Test approach → "Testing Strategy"
- Trade-offs → "Implementation Recommendations" (what to watch for)
6. User picks an interface (or accepts recommendation)
7. Draft and Create Task or Issue
Draft task or issue (depending on your context) using the template in TASK_TEMPLATE. Show the complete draft to the user with the question: "Create this task? (yes/no/edit)"
Only create the task/issue after explicit user approval. Share the URL or task id once created.
Prerequisites
- Codebase size: Minimum ~10 modules/files. Don't run on trivial codebases (< 500 LOC total) — the overhead isn't worth it.
- Test infrastructure: Some existing test coverage is required. If the codebase has 0% coverage, boundary tests won't help — establish basic testing practices first.
- Language support: Works best with languages that support interface definitions (TypeScript, Go, Rust, Scala, Java, C#). Scripting languages (Python, Ruby) work but require more discipline in interface design.
When to Skip This Skill
Don't use this skill if:
- Greenfield/prototype: The codebase is new and rapidly changing — wait until patterns stabilize.
- Config-heavy: Most "logic" is configuration wiring (YAML, JSON schemas) — there's nothing to deepen.
- Framework-locked: The architecture is dictated by a framework (Rails, Django, Next.js conventions) — work within the framework's patterns instead.
- No failing tests: If the codebase has no bugs and developers aren't complaining about navigation, don't invent problems.
- True external only: If all dependencies are third-party services (Stripe, AWS) with no internal logic to consolidate, deepening won't yield benefits.
Success Criteria
Stop and report completion when:
- No candidates found: After exploration, no shallow modules or tight coupling identified — report: "Codebase appears well-modularized."
- All candidates are True External: Dependencies are all third-party with no internal logic to consolidate — recommend dependency injection patterns instead.
- Task created: User-approved task/issue is created with the chosen interface design.
- User rejects all candidates: User decides none are worth pursuing — report why and close.
