Hooks are Python scripts triggered at Claude Code lifecycle events. They can inject context, block actions, or modify behavior.

All hooks live in .claude/hooks/ and share utilities from .claude/hooks/lib/config.py.

Skills are reusable instruction sets that activate when invoked.

Guides Claude through comprehensive planning so plans don't break during implementation:

1. Deep Discovery — Spawn Explore subagents to understand the codebase before writing anything
2. Design — Choose approach, define target state, verify assumptions against actual code
3. Decomposition — Break into subtasks with clear dependencies
4. Integration — Explicitly plan how modules connect (mandatory for multi-module plans)

Plans describe what and why, not how. The plan-reviewer agent validates plans against the actual codebase before implementation begins.

Creates and manages task folders. Each task has a context file (TASK-###-context.md) that serves as the single source of truth for that work:

• What was done and what's left
• Key decisions and why they were made
• Problems encountered and how they were solved
• Links to plan and related files

Why this matters: When you return to a task (or Claude resumes after compaction), the context file tells Claude exactly where things stand. No re-explaining, no lost progress.

Manages memory.jsonl via scripts (never edit manually):

# Add entry
python3 .claude/skills/memory-curator/scripts/add_memory_entry.py \
  --summary "Lesson learned about X" \
  --tags architecture,pattern \
  --links "TASK-042 src/service.ts"

# Edit entry
python3 .claude/skills/memory-curator/scripts/edit_memory_entry.py \
  --id mem-0042 --summary "Updated summary"

# Delete entry
python3 .claude/skills/memory-curator/scripts/delete_memory_entry.py \
  --id mem-0042

Agents are specialized subagents that validate work.

Validates plans before implementation:

• Verifies file paths and API assumptions against codebase
• Checks for missing steps, dependencies, integration plan
• Uses Context7 and DeepWiki to verify library claims
• Returns score (must reach 9+ to proceed) + findings

Validates implementations after completion:

• Compares actual code against plan
• Flags hardcoded values, TODOs, unused exports
• Checks integration (no orphaned modules)
• Returns score (must reach 9+ to ship) + findings

Multi-Reviewer Strategy: For large plans, spawn multiple focused reviewers in parallel — one per plan phase plus integration reviewer(s).

MCP (Model Context Protocol) servers give Claude access to up-to-date external knowledge. Meridian includes two:

Queries documentation for any public library. Claude uses this to:

• Verify library APIs exist and work as expected
• Find correct usage patterns and examples
• Check for deprecations or breaking changes

Asks questions about any public GitHub repository. Claude uses this to:

• Understand how external libraries work internally
• Verify integration patterns are correct
• Research best practices for specific tools

Why MCPs matter: Claude's training data has a cutoff date. When planning or reviewing, Claude can verify claims against current documentation instead of relying on potentially outdated knowledge.

Configuration: .mcp.json in project root:

{
  "mcpServers": {
    "context7": {
      "type": "http",
      "url": "https://mcp.context7.com/mcp"
    },
    "deepwiki": {
      "type": "http",
      "url": "https://mcp.deepwiki.com/mcp"
    }
  }
}

# Project type → which CODE_GUIDE addon to load
project_type: standard  # hackathon | standard | production

# TDD mode → tests first
tdd_mode: false

# Quality gates
plan_review_enabled: true
implementation_review_enabled: true

# Context preservation
pre_compaction_sync_enabled: true
pre_compaction_sync_threshold: 150000  # tokens

# Always injected
core:
  - .meridian/memory.jsonl
  - .meridian/task-backlog.yaml
  - .meridian/CODE_GUIDE.md
  - .meridian/prompts/agent-operating-manual.md

# Your custom docs (injected FIRST, before core files)
# Add architecture docs, API references, design specs — anything Claude should always know
user_provided_docs:
  - docs/architecture.md
  - docs/api-reference.md

# Auto-loaded based on project_type
project_type_addons:
  hackathon: .meridian/CODE_GUIDE_ADDON_HACKATHON.md
  production: .meridian/CODE_GUIDE_ADDON_PRODUCTION.md

user_provided_docs: Add any project-specific documentation here. These files are injected at the very top of context (before memory and core files), so Claude always has access to your architecture decisions, API contracts, or any other docs you want it to reference.

• Baseline (CODE_GUIDE.md) — Default standards for Next.js/React + Node/TS
• Hackathon Addon — Relaxes rules for fast prototypes
• Production Addon — Tightens rules for production systems
• TDD Addon — Overrides all testing rules with test-first workflow

Precedence: Baseline → Project Type Addon → TDD Addon