Claude Code Plugin
Disciplined Engineering Workflows
A lean toolkit that enforces TDD, systematic debugging, smart commits, code review, and knowledge capture across your engineering workflow.
The Ecosystem
Five tools, one cohesive workflow. The plugin is the heart — the rest amplify what you build.
What makes it different
Three principles that set Coding Friend apart from every other AI workflow tool.
Radical Simplicity
Install once. Start using immediately.
Most workflow tools require complex setup or hours of configuration. Coding Friend is the opposite — install once, use everywhere, with plain text skills you can read and edit anytime.
One install, every project
npm install once. The plugin loads into Claude Code automatically. No per-project configuration required.
Plain text skills
Skills are Markdown files. No DSL to learn. Read them, edit them, share them — they're just text.
Zero friction workflows
Type /cf-plan and start. Auto-invoked skills activate in the background — TDD, debugging, verification — without a thought.
Multi-platform out of the boxcoming soon
Works with Cursor, Windsurf, Copilot, Roo Code, and more. One toolkit, every AI coding tool.
/cf-learn
Your knowledge, amplified.
Vibe-coding is fast — but you risk learning nothing. Every session generates insights you never write down. /cf-learn captures them as human-readable docs, so you actually understand what was built — not just the AI.
$ /cf-learn JWT AuthenticationExtracting knowledge from session...📄 JWT Authentication Patterns├── Token validation flow├── Middleware composition└── Error handling strategiesSaved → docs/learn/Web_Dev/jwt-auth.md$ cf host # Browse as website$ cf mcp # Feed into any LLM
CF Memory
AI that remembers your project.
Every AI session starts from scratch — repeating the same mistakes, forgetting past decisions, losing debug knowledge. CF Memory gives your AI persistent, searchable memory across sessions — so it learns from your project over time.
$ cf memory statusTier 1 (SQLite + Hybrid Search)Memories: 47 │ Daemon: runningIndex: FTS5 + sqlite-vec (384d)$ cf memory search "auth middleware"┌─ 0.94 Auth middleware uses JWT httpOnly├─ 0.87 CORS fix: missing Origin header└─ 0.81 Session rotation on refreshSkills auto-recall relevant memoriesbefore every task via MCP.
3-tier graceful degradation
SQLite + vectors when available, MiniSearch in-memory as fallback, pure grep as baseline. Always works, even offline.
Hybrid search (BM25 + semantic)
Keyword precision meets semantic understanding. Find memories even when you paraphrase — RRF fusion ranks the best results.
Auto-capture from skills
cf-fix, cf-sys-debug, cf-review, and cf-ask automatically store debug episodes, architectural insights, and solutions.
100% local, markdown source of truth
No cloud, no API keys. Your memories live as plain markdown files you can read, edit, and version-control.
Workflow in action
Skills and agents work together automatically.
Plus utility skills: /cf-ask, /cf-remember, /cf-research, /cf-session, /cf-help
/cf-plan
Brainstorm your approach. Explores the codebase, identifies patterns, and writes a structured plan.
cf-tdd
TDD workflow enforced. Every feature starts with a failing test, then implementation, then refactor.
/cf-review
Multi-layer code review in a separate subagent. Checks security, performance, style, and coverage.
/cf-commit
Smart conventional commit. Analyzes staged changes and writes a meaningful commit message.
/cf-ship
One command to verify, commit, push, and create a pull request. Smart conventional commits included.
Memory Architecture
3-tier graceful degradation — always works, even without heavy dependencies. See Memory System for more details.
Your AI coding session. Skills like /cf-fix, /cf-ask, and /cf-remember trigger memory operations automatically.
Central routing hub. Detects the best available backend (SQLite → Daemon → Grep) and routes all 6 memory tools directly to it.
Full hybrid search with semantic vectors
In-memory index hosted by background daemon (Hono + UDS)
Zero deps, always available
Source of truth. Human-readable markdown with YAML frontmatter. Git-trackable, portable, never locked in.
Features
Beyond skills — the tools and systems that power your workflow.
Always know what's happening. Your terminal statusline, upgraded with real-time project context — version, branch, model, context usage, and rate limits at a glance. Read more about it here.
✻ | ▟█▙ Claude Code v2.1.50 ▐▛███▜▌ Opus 4.6 · Claude Max ▝▜█████▛▘ ~/git/coding-friend ▘▘ ▝▝
Run cf statusline to set up in seconds.
/cf-learn extracts human-readable knowledge from your coding sessions. Then cf host turns those docs into a searchable static website you can browse locally or deploy anywhere.
/cf-learnExtract knowledge
docs/learn/Markdown files
cf hostStatic website
Save your Claude Code session to a portable markdown file and resume it on any machine. Perfect for switching between desktop and laptop, or handing off work to a teammate.
/cf-sessionSave current session context, messages, and file changes to docs/sessions/.
cf session loadRestore a saved session on any machine — pick up exactly where you left off
$ /cf-sessionSaving current session...📋 Session Summary├── 47 messages, 3 files changed├── Context: auth middleware refactor└── Branch: feat/jwt-authSaved → docs/sessions/2026-03-17_jwt-auth.md$ cf session load docs/sessions/2026-03-17_jwt-auth.mdRestoring session context...✓ Session loaded — continue where you left off
Every AI session starts from scratch — repeating mistakes, forgetting decisions. CF Memory gives your AI persistent, searchable memory across sessions with 3-tier graceful degradation.
SQLite
FTS5 + semantic vectors with RRF fusion
MiniSearch
In-memory BM25 + fuzzy matching via daemon
Grep
Zero-dependency file scan, always available
Auto-capture from skills
cf-fix, cf-review, cf-ask automatically store debug episodes and architectural insights.
Hybrid search (BM25 + semantic)
Find memories even when you paraphrase — Reciprocal Rank Fusion ranks the best results.
100% local, markdown source of truth
No cloud, no API keys. Human-readable markdown with YAML frontmatter, git-trackable.
MCP integration
Skills auto-recall relevant memories before every task via the MCP server.
Skills are your commands — type /cf-plan and Coding Friend handles the rest. Behind the scenes, agents do the heavy lifting: specialized subprocesses that explore, plan, review, and write autonomously.
13 Slash Commands
Manual triggering with natural language
/cf-plan/cf-fix/cf-ask/cf-review/cf-commit/cf-ship/cf-optimize/cf-scan/cf-remember/cf-learn/cf-research/cf-session/cf-help4 Auto-Invoked Skills
Activate automatically when relevant — TDD, debugging, verification, code review
6 Specialized Agents
Autonomous subprocesses for parallel work
cf-explorerCodebase exploration (read-only, fast)cf-plannerTask decomposition & brainstormingcf-implementerTDD implementation in isolationcf-code-reviewerMulti-layer code reviewcf-writerLightweight doc generationcf-writer-deepDeep reasoning for nuanced contentSkills orchestrate agents behind the scenes — /cf-review dispatches cf-code-reviewer, /cf-plan uses cf-planner + cf-explorer.
/slash skills
Manual triggering with natural language. Coding Friend handles the rest.
/cf-planMedium~1K–2.5K tokens injected into promptPlan
Brainstorm and create implementation plans with structured exploration
/cf-fixMedium~1K–2.5K tokens injected into promptFix
Quick bug fix workflow with problem verification and approach confirmation
/cf-askMedium~1K–2.5K tokens injected into promptAsk
Quick Q&A about your codebase with persistent memory
/cf-optimizeMedium~1K–2.5K tokens injected into promptOptimize
Structured optimization with before/after measurement
/cf-reviewMedium~1K–2.5K tokens injected into promptReview
Multi-layer code review in a forked subagent
/cf-commitLow<1K tokens injected into promptCommit
Smart conventional commits with diff analysis
/cf-shipLow<1K tokens injected into promptShip
Verify, commit, push, and create PR in one command
/cf-scanHigh>2.5K tokens injected into promptScan
Scan project and bootstrap memory with architecture and conventions
/cf-rememberMedium~1K–2.5K tokens injected into promptRemember
Capture project knowledge for AI memory across sessions
/cf-learnMedium~1K–2.5K tokens injected into promptLearn
Extract human learning docs from vibe coding sessions
/cf-researchMedium~1K–2.5K tokens injected into promptResearch
In-depth research with web search and parallel subagents
/cf-sessionMedium~1K–2.5K tokens injected into promptSession
Save current session to docs/sessions/ to resume on another machine
/cf-helpMedium~1K–2.5K tokens injected into promptHelp
Answer questions about Coding Friend — skills, agents, workflows
Auto-invoked skills
These skills activate automatically when relevant. No slash needed.
cf-tddMedium~1K–2.5K tokens injected into promptTDD Workflow
Enforces test-driven development: RED, GREEN, REFACTOR
cf-sys-debugMedium~1K–2.5K tokens injected into promptSystematic Debug
4-phase debugging methodology for systematic problem solving
cf-auto-reviewLow<1K tokens injected into promptCode Review
Multi-layer review checklist automatically applied
cf-verificationLow<1K tokens injected into promptVerification Gate
Ensures tests pass before claiming work is done
Custom skill guides
Extend any built-in skill with your own rules. Add steps before, after, or throughout any workflow.
## BeforeRuns before the builtin workflow starts
## RulesAdditional rules applied throughout
## AfterRuns after the final step completes
Create .coding-friend/skills/<skill-name>-custom/SKILL.md to customize any skill.
Specialized agents
Skills dispatch work to focused agents, each optimized with the right model for their task.
cf-explorerLow<1K tokens injected into promptExplorer
Maps project structure, searches files, reads code, and returns structured reports with findings and dependencies.
cf-code-reviewerLow<1K tokens injected into promptCode Reviewer
Multi-layer review covering plan alignment, code quality, security (OWASP top 10), and test coverage.
cf-implementerLow<1K tokens injected into promptImplementer
Strict TDD implementation in an isolated context. RED, GREEN, REFACTOR — then reports back.
cf-plannerLow<1K tokens injected into promptPlanner
Designs 2-3 implementation approaches with pros, cons, effort, risk, and a recommended path.
cf-writerMedium~1K–2.5K tokens injected into promptWriter
Lightweight doc generation for memory files, notes, and straightforward markdown content.
cf-writer-deepLow<1K tokens injected into promptWriter Deep
Deep reasoning writer for nuanced technical content, complex trade-off analysis, and architecture docs.
Get started in 3 steps
Install once, use in every project. Supports user, project, and local scopes. More details in the documentation.
npm i -g coding-friend-clicf installcf init