Discover

Overview

Two modes of information gathering: codebase mapping (document what exists) and research (gather validated external knowledge).

Mode 1: Codebase Mapping

Purpose

Document the existing codebase objectively. You are a camera, not a critic.

Principles

• No opinions - Document what is, not what should be
• No recommendations - That's for design phase
• File:line references - Everything traceable
• Complete coverage - Architecture, patterns, constraints

Process

1. Identify scope (what part of codebase is relevant)
2. Map architecture and structure
3. Document patterns and conventions
4. Note integration points
5. List constraints and limitations

Output: codebase.md

# Codebase Analysis: [Feature Name]

**Date:** YYYY-MM-DD
**Scope:** [What was analyzed]

## Architecture Overview

[High-level structure with file:line references]

## Relevant Patterns

### [Pattern Name]

[Documentation with file:line references]

## Integration Points

[Where new code will connect]

## Constraints

[Limitations discovered]

Commit

After creating codebase.md:

explore: document codebase for [feature-name]

---

Mode 2: Research

Purpose

Gather validated knowledge from authoritative sources. Trust nothing from memory.

Principles

• Epistemic humility - Assume you know nothing
• Authoritative sources - Official docs, specs, RFCs
• Verify everything - No stale training data
• Document sources - Every claim has attribution
• Flag uncertainty - Honest about gaps

Process

1. Identify what needs researching (from codebase.md)
2. Fetch official documentation
3. Verify claims against authoritative sources
4. Cross-reference when possible
5. Document with full attribution

Output: research.md

# Research: [Feature Name]

**Date:** YYYY-MM-DD
**Technologies:** [List]

## Best Practices

### [Technology/Pattern]

**Source:** [URL]
**Retrieved:** YYYY-MM-DD

[Findings]

## Security Considerations

**Source:** [URL]

[Findings]

## Performance Guidance

**Source:** [URL]

[Findings]

## Sources

| Source | URL | Retrieved | Notes |
|--------|-----|-----------|-------|

Commit

After creating research.md:

explore: research [technology/pattern] best practices

---

What This Skill Does NOT Do

• Make recommendations (design phase)
• Critique existing code (you're a camera)
• Speculate or assume (verify everything)
• Use outdated training data (fetch current docs)