Workflows

Coding Friend installs globally. No files are copied into your project. On every Claude Code session, hooks auto-run: detect project type, inject rules, block sensitive files, and activate skills (TDD, debugging, verification, learn) based on context. You interact via slash commands.

Setup

• New project: run cf init to create workspace folders (docs/plans, docs/memory, docs/research, docs/learn), configure .gitignore, language, and /cf-learn settings.
• Existing project: nothing needed — Coding Friend is already active. Optionally run cf init for project-specific settings.
• Per-project config (optional): .coding-friend/config.json to customize docs folder, toggle hooks. .coding-friend/ignore to block directories from agent access.

Plan

/cf-plan Build a REST API with auth

Explores codebase and requirements, brainstorms 2-3 approaches, picks the best one, writes a plan to docs/plans/YYYY-MM-DD-<slug>.md, and creates a todo list.

Implement (auto)

TDD kicks in automatically when writing production code — no slash command needed.

1. RED — write a failing test first
2. GREEN — minimum code to pass
3. REFACTOR — clean up, all tests must still pass

Fix a Bug

/cf-fix Login fails with 401 when token is valid

Reproduce, locate root cause, fix, verify, and add a regression test. If 3 fix attempts fail, auto-escalates to cf-sys-debug (4-phase systematic debugging).

Review

/cf-review

Runs in a forked context so it doesn't pollute your main conversation.

• /cf-review — all uncommitted changes
• /cf-review src/auth/ — specific directory
• /cf-review HEAD~3..HEAD — last 3 commits

Reports: Critical (must fix) > Important (should fix) > Suggestions.

Commit

/cf-commit Add retry logic for API calls

Runs tests first, stages relevant files, writes a conventional commit message, and commits.

Ship

/cf-ship Add notifications

Full pipeline in one command: verify (tests, build, lint) > commit > push > create PR.

Quick Q&A

/cf-ask How does the auth middleware work?

Explores the codebase to find the answer, then saves Q&A to docs/memory/ for future sessions.

Optimize

/cf-optimize database query in getUserById

Baseline (3 runs) > analyze bottlenecks > plan > implement (TDD) > measure after > compare.

Research

/cf-research GraphQL vs REST for mobile APIs

Deep research with web search and parallel subagents. Structured output saved to docs/research/.

Save Knowledge

/cf-remember auth flow

Saves project knowledge for the AI to use in future sessions (docs/memory/). Run without arguments to scan the entire conversation for features, conventions, decisions, and gotchas.

Learn

/cf-learn dependency injection

Extracts learnings so you actually learn from vibe coding — not just let AI do everything. Also auto-invoked when substantial new knowledge is detected. Output to configured dir (default: docs/learn/). Read with cf host or use as MCP server with cf mcp.

Typical Flow

1. /cf-plan Add dark mode toggle — plan
2. Implement (TDD auto-enforced) — code
3. /cf-review — review
4. /cf-ship Dark mode feature — ship
5. /cf-remember + /cf-learn — capture knowledge

What's Automatic vs Manual

Automatic (no action needed): TDD enforcement, systematic debugging (after 3 failed fixes), verification before done, privacy block (.env, credentials), scout block (node_modules, dist), prompt injection defense, session bootstrap, rules injection, learning extraction on substantial knowledge.

Manual (slash commands): cf init, /cf-plan, /cf-fix, /cf-ask, /cf-optimize, /cf-review, /cf-commit, /cf-ship, /cf-remember, /cf-learn, /cf-research.