| name | exploration-explore-code |
| description | Use when user wants to understand unfamiliar code before extracting functionality - exploring database layers, API usage, library patterns, or any code to be extracted into a reusable component |
Explore Code
Deep exploration of unfamiliar codebases to understand usage patterns before extraction or integration.
Overview
This skill helps users understand code they don't know, specifically when preparing to extract functionality into reusable libraries or components.
MANDATORY: User must specify what usage type to explore (e.g., "database reading patterns", "authentication flow", "API client usage").
The exploration answers three questions:
- What is used - Which tables, APIs, components, functions
- How it is used - Patterns, conditions, parameters, transformations
- Where to use it - Call sites, contexts, triggering conditions
Key assumption: The project is unknown to the user. They're learning to extract, not verifying known behavior.
Output: A structured markdown file at planning/exploration-<topic>.md with precise usage inventory for extraction planning.
Process Flowchart
┌─────────────────────────────────────────────────────────────────┐
│ User asks exploration question │
└─────────────────────────────────────────────────────────────────┘
│
▼
┌───────────────────────┐
│ Ask scoping questions │◄────┐
└───────────────────────┘ │
│ │
▼ │
┌───────────────────────┐ │
│ Scope clear? │─────┘
└───────────────────────┘ no
│ yes
▼
┌───────────────────────┐
│ Initial exploration │
└───────────────────────┘
│
▼
┌───────────────────────┐
│ CHECKPOINT 1: Report │
│ Ask: Focus areas? │
└───────────────────────┘
│
▼
┌───────────────────────┐
│ Deep dive on areas │◄────┐
└───────────────────────┘ │
│ │
▼ │
┌───────────────────────┐ │
│ CHECKPOINT 2: Report │ │
│ Ask: Skip/add areas? │ │
└───────────────────────┘ │
│ │
▼ │
┌───────────────────────┐ │
│ More areas? │─────┘
└───────────────────────┘ yes
│ no
▼
┌───────────────────────┐
│ Synthesize findings │
└───────────────────────┘
│
▼
┌───────────────────────┐
│ CHECKPOINT 3: Draft │
│ Ask: Adjust format? │
└───────────────────────┘
│
▼
┌───────────────────────┐
│ Generate final doc │
└───────────────────────┘
│
▼
┌───────────────────────┐
│ Deliver document │
└───────────────────────┘
Phase 1: Scoping Questions
MANDATORY: Ask these BEFORE any exploration.
Use the AskUserQuestion tool to understand:
-
Usage type (REQUIRED): "What specific usage do you want to explore?"
- Examples: "database reading layer", "REST API client", "authentication flow", "file I/O patterns"
- This MUST be specified - do not proceed without it
-
Extraction goal: "What are you planning to extract or create?"
- Options: Standalone library, Internal module, API wrapper, Understanding only (no extraction)
-
Scope boundaries: "Which parts of the codebase are relevant?"
- Options: Specific directories, Specific file types, Everything, Not sure (I'll explore)
-
Depth: "How detailed should the inventory be?"
- Options: High-level patterns only, Detailed with all call sites, Exhaustive with parameters/conditions
Phase 2: Initial Exploration
After scoping, perform initial pass:
- Find relevant files using Glob patterns based on scope
- Identify key entry points - main files, exports, public APIs
- Map top-level structure - directories, modules, packages
CHECKPOINT 1: Present initial findings to user
## Initial Scan Results
**Files found:** N files matching criteria
**Key entry points:** [list]
**Main directories:** [list]
**Potential focus areas:**
1. [Area A] - appears to handle X
2. [Area B] - appears to handle Y
3. [Area C] - appears to handle Z
**Which areas should I focus on?** (can select multiple or suggest others)
Wait for user response before continuing.
Phase 3: Deep Exploration
For each selected area, answer the three core questions:
3.1 What Is Used
Inventory everything the code interacts with for the specified usage type:
- Classes, structs, functions accessed
- External libraries or APIs called
- Resources (files, connections, handles)
- Data structures and their members
3.2 How It Is Used
Document the patterns and conditions:
- Function signatures and parameter patterns
- Return value handling (error codes, exceptions, optionals)
- Conditional logic (when is X called vs Y?)
- Resource lifecycle (RAII patterns, manual management)
- Configuration and parameterization
3.3 Where To Use It
Map call sites and triggering contexts:
- Which functions/methods initiate the usage?
- What's the call hierarchy? (entry point → ... → actual usage)
- What conditions trigger each pattern?
- Lifecycle timing (initialization, request handling, shutdown)
CHECKPOINT 2: Present findings for each area
## Deep Dive: [Area Name]
### What Is Used
| Component | Type | Purpose |
|-----------|------|---------|
| DataReader | class | Reads records from source |
| Config::timeout | member | Connection timeout in ms |
| fetch() | method | Retrieves data with filters |
### How It Is Used
| Pattern | Signature/Example | Conditions |
|---------|-------------------|------------|
| Filtered fetch | fetch(Filter f) | When user specifies criteria |
| Batch read | fetchBatch(size_t n) | For bulk operations |
| Retry on error | 3 retries with backoff | On connection failure |
### Where To Use It
| Call Site | Context | Trigger |
|-----------|---------|---------|
| Processor::run() | Main loop | Each iteration |
| Importer::load() | Startup | Once on init |
**Should I continue to next area, go deeper here, or adjust scope?**
Wait for user response.
Phase 4: Synthesis
After all areas explored:
CHECKPOINT 3: Present draft document structure
## Proposed Document Structure
1. **What Is Used** (Complete inventory)
- [components to include]
2. **How It Is Used** (Patterns and conditions)
- [patterns to document]
3. **Where To Use It** (Call sites and contexts)
- [call hierarchies to include]
4. **Extraction Recommendations**
- [interface suggestions, dependencies to handle]
**Adjustments needed?** (add sections, remove sections, change focus)
Wait for user approval before generating.
Phase 5: Generate Document
Create the final markdown file with:
Required Sections
1. What Is Used
Complete inventory of components for the explored usage type:
| Component | Type | Purpose | Dependencies |
|---|
| ClassName | class | Brief purpose | lib1, lib2 |
| functionName() | function | Brief purpose | ClassName |
2. How It Is Used
Patterns, signatures, and conditions:
| Pattern | Signature/Example | When Used | Notes |
|---|
| Pattern name | func(param1, param2) | Condition | Gotchas |
3. Where To Use It
Call sites and triggering contexts:
| Entry Point | Call Chain | Trigger | Frequency |
|---|
| main() | → App::run() → Service::fetch() | On request | Per-request |
4. Extraction Recommendations
- Suggested interface for extracted library
- Dependencies to bundle or abstract
- Breaking changes to anticipate
- Specific answers to user's original question
Output File Location
Default: planning/exploration-<topic>.md
This integrates with the planning workflow - exploration documents feed into planning-feature or planning-refactor skills.
Subagent Usage
For large codebases, use subagents for parallel exploration:
- Spawn separate subagents for independent areas
- Each subagent reports back with What/How/Where for its area
- Main agent synthesizes findings
Example: "Explore database layer" might spawn:
- Subagent 1: Read operations
- Subagent 2: Write operations
- Subagent 3: Connection management
Main Agent
│
├── Subagent: "Explore read operations"
│ └── Returns: What/How/Where for reads
│
├── Subagent: "Explore write operations"
│ └── Returns: What/How/Where for writes
│
└── Subagent: "Explore connection management"
└── Returns: What/How/Where for connections
Main Agent synthesizes into single document
Red Flags - STOP and Check
- No usage type specified - User MUST specify what to explore (e.g., "database reading", "logging layer")
- Exploring without scoping - Go back and ask questions first
- Long silence - If you've read more than 5 files without user update, pause and report
- Missing what/how/where - Every finding should answer at least one of the three questions
- Generating without checkpoint - Always present draft structure before final document
Common Mistakes
| Mistake | Fix |
|---|
| Starting without usage type | Ask: "What specific usage do you want to explore?" |
| Diving into code without asking scope | Always run Phase 1 scoping questions first |
| Listing "what" without "how" or "where" | Every component needs usage patterns and call sites |
| Producing final doc without checkpoints | Stop at each checkpoint, wait for user |
| Prose-heavy output | Use tables, lists, diagrams - scannable format |
| Missing call hierarchies | Trace from entry point to actual usage |
Rationalizations to Reject
| Excuse | Reality |
|---|
| "User seems busy, I'll skip scoping" | Scoping saves time. Ask the questions. |
| "Usage type is obvious from context" | Confirm anyway. "Database" could mean read/write/schema/migrations. |
| "I found what's used, that's enough" | Without how/where, user can't extract safely. Complete all three. |
| "Checkpoints slow things down" | Checkpoints prevent wasted work. Do them. |
| "I'll figure out call sites later" | Call sites ARE the extraction boundary. Find them now. |
| "The code is simple, no doc needed" | User is unfamiliar with it. Document everything found. |
Integration with Other Skills
| After Exploration | Use This Skill |
|---|
| Extract as new library | planning-feature |
| Refactor existing code | planning-refactor |
| Just needed understanding | Done - no further action |
The exploration document (planning/exploration-<topic>.md) serves as input to planning skills.