Lean toolkit for disciplined engineering workflows with Claude Code.

Website · Documentation · Changelog · Report Bug

• Enforces test-driven development (TDD)
• Provides systematic debugging methodology
• Quick bug fix workflow (/cf-fix)
• Structured optimization with before/after measurement (/cf-optimize)
• Quick Q&A about codebase with memory (/cf-ask)
• Ensures verification before claiming done
• Smart conventional commits and code review
• Captures project knowledge across sessions (/cf-remember)
• ✨ Persistent AI memory with 3-tier hybrid search (cf memory) — stores facts, preferences, debug episodes across sessions with automatic recall
• ✨ Helps humans learn from vibe coding sessions (/cf-learn) — host as a local website (cf host) or setup MCP server (cf mcp) for other LLM clients
• In-depth research with web search and parallel subagents (/cf-research)
• Custom skill guides — extend built-in skills with your own Before/Rules/After per skill
• ✨ Save and load Claude Code session chats across machines and accounts (/cf-session)
• Prompt injection defense — layered content isolation protects against malicious instructions
• ✨ Customizable Claude Code statusline with API rate limit tracking

  cf v0.3.0 | 📂 MyProject (⎇ main) | 🧠 Opus (1M)
  ctx 42% | [5h] 30% → 2:30pm | [7d] 10% → mar 15, 2:30pm
  

For full details, visit the official documentation.

Main workflow:

Memory architecture (read more about it here):

The Learning Notes taken from your coding sessions with the help of the /cf-learn skill and cf host command:

# npm
npm i -g coding-friend-cli

# yarn
yarn global add coding-friend-cli

# pnpm
pnpm add -g coding-friend-cli

cf install              # Interactive — asks which scope to use
cf install --user       # All projects (default)
cf install --project    # This project only (shared via git)
cf install --local      # This machine only (gitignored)

Supports user, project, and local scopes. All features work identically across scopes.

cf init

This sets up workspace folders (docs/plans, docs/memory, docs/research, docs/learn) and optionally adds them to .gitignore.

Restart to load the plugin, then use slash commands like /cf-plan, /cf-fix, /cf-commit, etc.

The memory system stores project knowledge (facts, conventions, debug episodes) and recalls them automatically in future sessions. Basic memory works immediately, but you can enable better search:

cf memory start-daemon  # Start daemon with fuzzy search (Tier 2)
cf memory init       # Initialize SQLite with hybrid search (Tier 1)
cf memory status     # Check current tier and document count

Then bootstrap memory with project knowledge inside Claude Code:

/cf-scan This is a Next.js app with PostgreSQL and Stripe

This scans your project and creates ~10-15 memories covering architecture, conventions, and key features. Safe to run multiple times.

Learn more: cf memory, Memory System.

The /cf-learn skill generates learning notes from your coding sessions. You can browse them as a website or expose them to other LLM clients:

cf host              # Serve docs/learn/ as a website at localhost:3333
cf mcp               # Setup an MCP server so other LLM clients can read your notes

Learn more: cf host, cf mcp.

Auto-invoked skills (no slash needed): cf-tdd, cf-sys-debug, cf-auto-review, cf-verification.

The plugin is managed by the CLI cf command. Learn more about the CLI in the CLI documentation.

For plugin developers, check plugin-dev.md.

Read the official documentation.

MIT