Gave my Claude a subconscious memory system

An operating system for Claude Code. Skills, hooks, and a framework for making Claude Code sessions smarter, leaner, and persistent. Born from hundreds of hours of real production use. Every anti-pattern documented here destroyed real work at least once before it was caught. | Layer | What | Where | |---|---|---| | Hooks | Infrastructure that runs silently on every interaction | hooks/ | | Skills | Slash commands for orchestrating complex workflows | skills/ | | Framework | Agent execution model, file ownership, conventions | framework/ | A "subconscious brain" for Claude Code. Hooks that automatically inject relevant knowledge from a personal vault into every conversation — without Claude searching for it. Two layers: - Layer 1: UserPromptSubmit hook matches your prompt against a knowledge vault and injects context before Claude sees your message - Layer 2: PreToolUse hook intercepts tool calls (Read, Grep, WebSearch) as need signals and injects relevant vault context alongside tool results Things I learned building with Claude Code — informal patterns and observations from building a production app with these tools. Start here if you're new. The core idea is simple: your main chat agent is a team lead, not a worker. Every task — coding, research, exploration, testing, even reading files — gets delegated to subagents or agent teams. The main agent's context window stays lean, which means: - You can run longer sessions without hitting context limits - You can chain multiple skills in one session ( /brainstorm →/plan-features →/exec-plan ) - Each delegated agent gets a fresh, focused context window The framework lives in framework/ and includes: CLAUDE.md — Drop-in template for your project's.claude/CLAUDE.md . Includes the agent execution model, file ownership template, and code conventions template.rules/agent-execution.md — Auto-loaded rule that enforces lean-context discipline on every conversation.rules/file-ownership.md — Template for mapping directories to team names, preventing merge conflicts during parallel execution. | Skill | Command | What It Does | |---|---|---| | Brainstorm | /brainstorm | Guided ideation with feasibility research agents. Turns rough ideas into validated requirements docs. | | Debate | /debate | 5-agent adversarial evaluation. Critics attack, defenders respond, moderator writes verdict. | | Launch Review | /launch-review | Full pipeline: debate -> plan -> execute. Evaluates readiness, plans fixes, executes them. | | Plan Features | /plan-features | Generates execution plans with file conflict tables, dependency graphs, and per-team specs. | | Execute Plan | /exec-plan | State machine that reads a plan and mechanically executes it with worktree agents, commit gates, and test validation. | | Skill | Command | What It Does | |---|---|---| | Hygiene Audit | /hygiene-audit | 5-agent codebase analysis: structure, dependencies, duplication, splitting patterns, optimization research. | | Review PR | /review-pr | Reviews a GitHub PR for correctness, convention compliance, test coverage, and security. | | Full Audit | /audit | 8-10 parallel agents cross-referencing docs vs code vs tests vs infra. Produces severity-scored report with optional auto-fix pipeline. | | Dep Audit | /dep-audit | Security vulnerability scan and dependency freshness check. | | Skill | Command | What It Does | |---|---|---| | Fix Issue | /fix-issue | End-to-end GitHub issue resolution: read issue, implement fix, write tests, commit. | | Test All | /test-all | Run the full CI pipeline locally (lint, types, tests) in CI order. | | New Endpoint | /new-endpoint | Scaffold a new REST API endpoint following existing patterns. | | New MCP Tool | /new-mcp-tool | Add a new tool to a FastMCP server with proper descriptions and tests. | | DB Migration | /db-migration | Guide through database schema changes with backup, migration, and testing. | | Doc Architecture | /doc-architecture | Update architecture docs after code changes to keep docs in sync. | | Skill | Command | What It Does | |---|---|---| | Deploy | /deploy | Pre-deploy safety checks, push to trigger CI/CD, monitor deployment. | | Deploy Status | /deploy-status | Health check, recent CI runs, deployment status summary. | | Incident | /incident | Emergency response guide: assess, contain, diagnose, remediate, document. | | Skill | Command | What It Does | |---|---|---| | API Conventions | /api-conventions | Auth patterns, route organization, error responses, MCP tools, middleware. | | Dashboard Conventions | /dashboard-conventions | Frontend patterns, theming, API integration, file organization. | | Skill | Command | What It Does | |---|---|---| | Dream | /dream | Memory consolidation. Scans session transcripts for corrections, decisions, and preferences, then merges into persistent memory files. | Copy individual skill directories into your project's .claude/skills/ directory: # Copy a single skill cp -r skills/workflow/brainstorm /path/to/project/.claude/skills/