Skill: context-prime

Purpose

Prime Claude with deep repository context before starting any significant task. Ensures Claude has a full mental model of the codebase architecture, conventions, and constraints before writing a single line of code.

Trigger

Use at the start of:

• A new coding session on an unfamiliar area
• A complex feature that touches many files
• A debugging session for a hard-to-reproduce bug
• Onboarding to a new repository

Process

Step 1: Read Core Docs

In order:

1. CLAUDE.md — primary instructions and conventions
2. README.md — project overview and purpose
3. TOOLS.md — available tooling
4. agent/CLAUDE.md — agent-specific context (if exists)
5. Any docs/ folder overview files

Step 2: Map the Architecture

Scan the top-level directory structure. For each major directory:

• Identify its purpose
• Note key files within it
• Understand how it connects to other directories

Build a mental map: [module] → [responsibility] → [interfaces with]

Step 3: Find Conventions

Look for patterns across 5-10 representative files:

• Naming conventions (files, functions, classes, variables)
• Import organization style
• Error handling patterns
• Testing patterns (test file location, naming, fixtures)
• Comment/docstring style
• Type annotation usage

Step 4: Understand Data Flow

Trace the main data flow through the system:

• Entry points (API routes, CLI commands, event handlers)
• Core processing logic
• Storage/persistence layer
• Output/response formation

Step 5: Identify Constraints

Note any explicit constraints from CLAUDE.md or comments:

• Performance-sensitive areas
• Security boundaries
• Deprecated patterns to avoid
• In-progress refactors to be aware of

Step 6: Declare Readiness

Output a structured context summary:

## Context Prime Complete

### Project
[1-2 sentence description of what this project does]

### Architecture
- [Layer/Module]: [responsibility]
- [Layer/Module]: [responsibility]

### Key Conventions
- [Convention]: [example]
- [Convention]: [example]

### Data Flow
[Entry] → [Processing] → [Storage] → [Output]

### Constraints to Respect
- [Constraint]
- [Constraint]

### Ready for Task
I have sufficient context to begin. Proceeding with: [task description]

Rules

• Do NOT start implementing before completing this skill
• If CLAUDE.md contradicts README.md, CLAUDE.md wins
• Note areas of uncertainty explicitly rather than assuming
• Re-prime if switching to a completely different part of the codebase
• Keep the summary concise — this is a mental model, not a transcript

Value

This skill prevents the most common failure mode: writing code that technically works but violates project conventions, misunderstands architecture, or duplicates existing functionality.