---

name: context-explorer description: | Systematically explore codebases using search and LSP to understand code structure, intent, and behavior. READ_ONLY mode - no file modifications.

Context Explorer

Explore codebases using search and LSP to understand code structure and create complete specifications.

MODE: READ_ONLY - Cannot modify files.

Instructions

Prerequisites

• Access to codebase files
• Search tools (Grep, Glob)
• LSP navigation
• Understanding of 14-slot completeness model

Workflow

1. Scope: Define focus area and search terms
2. Discover: Find relevant files using Glob and Grep
3. Extract: Analyze tests, types, and docs
4. Verify: Cross-reference evidence
5. Regress: Check completeness against 14-slot model
6. Plan: Generate probes for gaps
7. Emit: Produce specification with evidence

Core Skills

This skill orchestrates these sub-skills:

1. codebase-search: Find files and patterns
2. type-analysis: Analyze type definitions
3. test-analysis: Extract behavior from tests
4. end-state-spec: Synthesize specification
5. acceptance-criteria: Extract testable criteria
6. risk-gaps: Identify risks and gaps
7. missing-details-regression: Check completeness
8. evidence-plan: Generate probe plan

Examples

Example 1: Understand a Feature

CopyUser: "How does authentication work?"

Agent:
1. Glob: **/auth/**/*.py
2. Grep: "authenticate|login|token"
3. Read test files for expected behavior
4. Read type definitions for contracts
5. Check completeness: 10/14 slots filled
6. Generate probes for gaps

Output: Specification with evidence

Example 2: Find Implementation

CopyUser: "Where is the user service implemented?"

Agent:
1. Grep: "class.*UserService"
2. LSP: Go to definition
3. LSP: Find references
4. Read related tests
5. Document interfaces

Output: Implementation map with dependencies

14-Slot Completeness Model

Status Values: FILLED | EMPTY | VAGUE | CONFLICTING

Guidelines

• READ_ONLY: Never modify files
• Start with tests and types (highest signal)
• Use LSP for navigation
• Check completeness against 14-slot model
• Generate specific probes, not vague questions
• Track confidence levels
• Flag conflicts explicitly

Evidence Priority

Probe Types