Answer questions about Coding Friend — skills, agents, hooks, CLI, config, memory, hosts, parameters. Auto-invoke when the user asks about Coding Friend capabilities, available skills, how to use a skill, what skills / agents / commands / hooks exist, flags or parameters, how it works, or how to configure it — e.g. "what skills are available?", "how does coding friend work?", "what can you do?", "list all skills", "what agents exist?", "how do I use cf-plan?", "what is cf-tdd?", "what flags does X have?", "what hooks exist?", "how do I get started?", "coding friend features/setup", "does X require the CLI?", "what works without coding-friend-cli?", "is the CLI required?". Do NOT auto-invoke for general coding questions unrelated to Coding Friend itself.
Answer questions about Coding Friend — skills, agents, hooks, CLI, config, memory, hosts, parameters. Auto-invoke when the user asks about Coding Friend capabilities, available skills, how to use a skill, what skills / agents / commands / hooks exist, flags or parameters, how it works, or how to configure it — e.g. "what skills are available?", "how does coding friend work?", "what can you do?", "list all skills", "what agents exist?", "how do I use cf-plan?", "what is cf-tdd?", "what flags does X have?", "what hooks exist?", "how do I get started?", "coding friend features/setup", "does X require the CLI?", "what works without coding-friend-cli?", "is the CLI required?". Do NOT auto-invoke for general coding questions unrelated to Coding Friend itself.
user-invocable
true
model
haiku
allowed-tools
["Read","Glob","Grep"]
created
2026-02-17T00:00:00.000Z
updated
2026-08-26T00:00:00.000Z
/cf-help — Coding Friend Help
CLI Requirement: NONE — Works without coding-friend-cli. See CLI requirements for the full matrix.
Answer questions about the Coding Friend toolkit. You are the single place to ask anything about CF — skills, agents, hooks, CLI, config, memory, hosts, flags. Catalog below covers "what exists". For how / flags / config, follow the lookup rule — never guess.
Workflow
Step 0: Custom Guide
Custom guide — auto-loaded below (if the raw command shows instead of its output, run it yourself):
Memory / MCP — tiers, tools, /cf-remember vs /cf-learn
Workflow — how do skills work together?
Troubleshooting — skill not triggering, CLI, hooks, MCP
Lookup rule (do not skip)
Never invent a flag, config key, hook name, CLI command, or host difference.
A "what exists / which skill should I use" question → answer from the catalog below.
Anything more specific → Read ${CLAUDE_PLUGIN_ROOT}/skills/cf-help/topics.md (or plugin/skills/cf-help/topics.md in this repo) — topic → source map, flags, hooks, CLI, config.
Then Read the source file that index points to (skills/<name>/SKILL.md, agents/<name>.md, hooks/<file>, project config).
If still unclear: Glob / Grep under ${CLAUDE_PLUGIN_ROOT}/skills/, agents/, hooks/.
Step 2: Provide overview (if general question)
Coding Friend is a lean toolkit for disciplined engineering workflows in Claude Code. Core philosophy:
Check skills first — Before any task, check if a relevant skill exists
Test before code — RED → GREEN → REFACTOR
Verify before claiming — Never claim done without running tests
Commit with purpose — Conventional commits with clear "why"
Supported hosts: Claude Code (default), Codex CLI (beta), omp (beta), and Google Antigravity (beta) (--agent agy / --agy).
Two packages: the plugin (skills / agents / hooks — works alone) and optional coding-friend-cli (memory MCP, learn-host, cf init / install / config). Skills never call the cf binary.
Slash Commands (user triggers with /)
/cf-advise [decision] — ⚡⚡ — Decision advisory: a structured interview (one question at a time) that surfaces hidden requirements, confirms a reframing, then delivers a verdict-first recommendation with pitfalls and ranked alternatives. Advisory-only — never writes code or plans (that's /cf-plan). Flags: --quick fewer questions, --save persist the decision to docs/memory/decisions/.
/cf-ask [question] — ⚡⚡ — Quick Q&A about codebase → docs/memory/; auto-generates an ASCII flow diagram for "how does X work" / flow / lifecycle questions
/cf-plan [task] — ⚡⚡ — Brainstorm and create phased implementation plans with parallel execution. Flags: --fast (alias --quick) lighter workflow, --hard deeper exploration + rollback, --auto end-to-end autopilot (auto review + fix Critical/Important + commit per phase), --inline (alias --no-file) plan in chat only without writing a file, --gui (alias --human) also generate the human-readable overview doc for this run (off by default), --model <alias> pin the model for cf-planner at the brainstorm step.
/cf-plan-resume <plan> — ⚡⚡ — Resume a saved plan (folder path, entry file, or bare <slug>) from where execution last stopped: reads the plan + its context file, re-runs pending/interrupted tasks, honors auto: true frontmatter to continue in autopilot.
/cf-later-do [item] — ⚡⚡ — Work through deferred side-tasks in docs/later/: list captured items, pick one, route the fix to /cf-fix (bugs) or /cf-plan (features), remove the file only after the fix is verified-done, then suggest the next. The read/resolve side of capture-later.sh.
/cf-review [target] — ⚡⚡ — Dispatch code review to subagent. Flags: --with-codex/--codex, --claude, --gemini, --cursor, --grok run headless external reviewers in parallel and merge into one report; --out exports a /cf-review-out prompt with Claude's findings embedded. Set review.withCodex: true in config to enable Codex by default; review.agentTimeout (default 300s) bounds each external agent. Unavailable agents are skipped with a warning.
/cf-review-out [label] — ⚡⚡ — Write a self-contained review prompt + diff to docs/reviews/ for an external AI or human. Pair with /cf-review-in. Prefer /cf-review --claude|--gemini|--cursor|--grok|--codex when those CLIs are installed.
/cf-review-in <label> [service] — ⚡⚡ — Read an external review result, present findings, offer to fix.
/cf-teach [topic] — ⚡⚡ — Conversational teacher narrative of what just happened and why (human understanding). Unlike /cf-learn (structured notes).
/cf-research [topic] — ⚡⚡ — In-depth research with web search → docs/research/
/cf-session [label] — ⚡⚡ — Save current session to sync folder for cross-machine resume
/cf-warm [user] — ⚡⚡ — Catch up after absence — git history summary for a user
/cf-checkpoint [additional-prompt] — ⚡⚡ — Capture a concise conversation checkpoint (decisions, breaking changes, next steps) → docs/context/checkpoints/; updates an existing checkpoint when the arg matches one.
/cf-checkpoint-from <slug> [message] — ⚡⚡ — Load a saved checkpoint as context, then do the message (first word is the slug, the rest is what to do next). Add --recap to also print a summary of the restored context.
/cf-help [question] — ⚡⚡⚡ — This skill — answer questions about Coding Friend. Also auto-invoked.
Auto-Invoked Skills (activate automatically when relevant)
cf-tdd — ⚡⚡ — When writing new code: direct implementation by default; TDD with --add-tests or config tdd: true. Add --auto for standalone autopilot (auto review + fix + commit after implementation).
cf-sys-debug — ⚡⚡ — When debugging: investigate → analyze → test → fix
cf-reviewer-reducer (haiku) — Deduplicates and ranks findings
cf-implementer — ⚡ — Implementation subagent: direct coding by default, TDD with --add-tests (reads structured context file, returns result signals, supports auto-retry on failure). Does not own autopilot loops — cf-plan / cf-tdd orchestrate review / fix / commit when --auto is active.
cf-explorer — ⚡ — Codebase exploration and context gathering (writes structured context files for downstream agents)
Each skill loads its SKILL.md into context when triggered. Context tiers: ⚡ = low (<1,500 tokens), ⚡⚡ = medium (1,500–3,000), ⚡⚡⚡ = high (>3,000). Bootstrap context (~2,100 tokens) is loaded every session. Agents run in forked sessions with their own context window. For exact token counts, see https://cf.dinhanhthi.com/docs/reference/context-usage/.
Step 3: Read specific files (if detailed question)
Follow the lookup rule. Typical paths (plugin root = ${CLAUDE_PLUGIN_ROOT}):
skills/cf-help/topics.md ← always start here for details
skills/<skill-name>/SKILL.md
agents/<agent-name>.md
hooks/hooks.json
hooks/<hook-file>
Skill not triggering? Check description in SKILL.md — it may not match the user's phrasing. Use /cf-<skill-name> to trigger manually.
Custom guide not loading? Verify the path: .coding-friend/skills/<skill-name>-custom/SKILL.md and that it has ## Before, ## Rules, or ## After sections.
Config not applied? Local .coding-friend/config.json overrides global ~/.coding-friend/config.json. Check both.
After editing plugin files? Run cf dev sync to copy changes to the cached version.
More issues? Point the user to the Troubleshooting page for memory daemon, install, hook, and MCP issues.
Step 6: Answer concisely
Provide a clear, concise answer based on the information gathered. Link to specific files if the user wants to dive deeper. If you did not read a source file for a specific-skill / flag / hook / CLI / config question, go back to the lookup rule first.
CLI Requirements (quick reference)
The Coding Friend plugin works without coding-friend-cli. The CLI adds the memory MCP server (fast indexed search), the learn-host doc server, and a few utilities — but every skill and agent has a documented fallback path.
Three tiers:
NONE — works with zero CLI involvement.
OPTIONAL — uses CLI-installed memory MCP for speed; falls back to grep over docs/memory/ and direct file writes when CLI is absent. Full functionality preserved.
REQUIRED — cannot function without CLI. (0 skills today.)
When users ask whether a skill needs the CLI, look up its tier first.
Q: "Do I need the CLI to use /cf-fix?"
A: "No. cf-fix is OPTIONAL-tier — it uses the memory MCP when available, but falls back to grep -r '<query>' docs/memory/. See docs/cli-requirements.md for the full matrix."
