$cf-help — Coding Friend Help
Catalog = what exists. Flags / config / hooks → lookup rule. Never guess.
Workflow
Step 0: Custom Guide
bash "${PLUGIN_ROOT}/lib/load-custom-guide.sh" cf-help
If the block above printed anything, apply only the ## Before, ## Rules, and ## After sections; if it shows the raw command instead of output, re-run that exact load-custom-guide.sh fence now.
Step 1: Understand the question
Classify (more than one is fine): general overview · specific skill · specific agent · hooks / fewer prompts · CLI · setup / config · memory / MCP · workflow · troubleshooting.
Lookup rule (do not skip)
Never invent a flag, config key, hook name, CLI command, or host difference.
- "What exists / which skill?" → catalog below.
- Anything more specific → Read
${PLUGIN_ROOT}/skills/cf-help/topics.md (or plugin/skills/cf-help/topics.md in this repo) — topic → source map, flags, hooks, CLI, config, native prompt reduction.
- Then Read the source that index points to (
skills/<name>/SKILL.md, agents/<name>.md, hooks/<file>).
- If still unclear: Glob / Grep under
${PLUGIN_ROOT}/skills/, agents/, hooks/.
Step 2: Provide overview (if general question)
Coding Friend is a lean toolkit for disciplined engineering workflows in Codex CLI. Check skills first; TDD optional (--add-tests or tdd: true); verify before claiming done.
Hosts: Claude Code (default), Codex CLI, omp, Google Antigravity (--agent agy / --agy). Plugin works alone; coding-friend-cli is optional. Skills never call cf. Flags / config / native prompt-reduction: topics.md.
Slash Commands (user triggers with /)
$cf-advise [decision] — ⚡⚡ — Advisory interview. --quick, --save$cf-ask [question] — ⚡⚡ — Codebase Q&A → docs/memory/$cf-plan [task] — ⚡⚡ — Phased plans. --fast/--quick, --hard, --auto, --inline/--no-file, --gui/--human, --model <name> pin the model for cf-planner at the brainstorm step (Codex model name, e.g. gpt-5.5; Claude aliases are not valid). Autopilot review-fix cap: review.maxRounds (default 5). Normal/hard mode also writes brief.md (request, Q&A, assumptions) for $cf-plan-review.$cf-plan-resume <plan> — ⚡⚡ — Resume a saved plan. Honors auto: true.$cf-plan-review [plan] — ⚡⚡ — Review a saved plan with a fresh in-session reviewer before implementing. --codex, --gemini, --claude, --cursor, --grok add external reviewers in parallel (the flag matching the current host is skipped). Writes review.md into the plan folder, offers to apply Critical/Important findings.$cf-later-do [item] — ⚡⚡ — Resolve docs/later/ via $cf-fix or $cf-plan$cf-review [target] — ⚡⚡ — Dispatch review. Flags: --claude, --gemini, --cursor, --grok run headless external reviewers in parallel and merge into one report; --out exports a $cf-review-out prompt with in-session findings embedded. --with-codex/--codex and review.withCodex are ignored on Codex (do not spawn a nested Codex review). review.agentTimeout (default 300s) bounds each external agent; review.maxRounds (default 5) caps the autopilot fix loop. Unavailable agents are skipped with a warning.$cf-review-out [label] — ⚡⚡ — Prompt + diff → docs/reviews/$cf-review-in <label> [service] — ⚡⚡ — Read external review, offer to fix$cf-commit [hint] — ⚡ — Conventional commit$cf-design [mode] — ⚡⚡ — Scan / design / modify UI$cf-ship [hint] — ⚡ — Verify, commit, push, PR (--dry-run)$cf-fix [bug] — ⚡⚡ — Quick bug-fix$cf-optimize [target] — ⚡⚡ — Baseline → optimize → measure$cf-scan [description] — ⚡⚡⚡ — Bootstrap memory$cf-remember [topic] — ⚡⚡ — Project knowledge → docs/memory/. Also auto-invoked.$cf-learn [topic] — ⚡⚡ — Educational notes. Also auto-invoked.$cf-teach [topic] — ⚡⚡ — Story of what happened$cf-research [topic] — ⚡⚡ — Web research → docs/research/$cf-session [label] — ⚡⚡ — Cross-machine session save$cf-warm [user] — ⚡⚡ — Git history catch-up$cf-checkpoint [additional-prompt] — ⚡⚡ — Checkpoint → docs/context/checkpoints/$cf-checkpoint-from <slug> [message] — ⚡⚡ — Restore checkpoint, then act. --recap$cf-help [question] — ⚡⚡⚡ — This skill. Also auto-invoked.
Auto-Invoked Skills (activate automatically when relevant)
- cf-tdd — ⚡⚡ — Code-writing gate. Default direct; TDD with
--add-tests or tdd: true. --auto = review + fix + commit (review.maxRounds, default 5).
- cf-sys-debug — ⚡⚡ — 4-phase debug
- cf-verification — ⚡ — Evidence gate before claiming done
- cf-learn / cf-remember / cf-help — also auto-invoked (see above)
Agents (run in forked sessions — separate context window)
- cf-reviewer — ⚡ — Orchestrator: cf-reviewer-plan (medium reasoning effort), cf-reviewer-security (medium reasoning effort), cf-reviewer-quality (low reasoning effort), cf-reviewer-tests (low reasoning effort), cf-reviewer-rules (low reasoning effort), cf-reviewer-reducer (low reasoning effort)
- cf-implementer — ⚡ — Writes code; TDD with
--add-tests. [CF-RESULT: success|failure]. No autopilot.
- cf-explorer — ⚡ — Repo map + context files
- cf-planner — ⚡ — Approaches + phased tasks
- cf-writer — ⚡ — Straightforward markdown
- cf-writer-deep — ⚡ — Nuanced technical docs
Hooks (automatic — not slash commands)
hooks/hooks.json (+ *.agy.*; Codex transformed). Keys + native modes: topics.md.
session-init.sh · rules-reminder.sh · privacy-block.sh · scout-block.cjs · auto-approve.cjs (autoApprove; Claude autoApproveLLM default false → unknown defers to native) · session-log.sh · task-tracker.sh · agent-tracker.sh · memory-capture.sh · statusline.sh (Claude; cf statusline)
CLI (coding-friend-cli, binary cf) — optional
Lifecycle install|uninstall|enable|disable|update (--user|--project|--local, --agent claude|codex|omp|agy); setup init config permission statusline; memory / learn / mcp; status clean session guide dev. No skill requires the CLI. Full flags: topics.md.
Tiers ⚡ / ⚡⚡ / ⚡⚡⚡: https://cf.dinhanhthi.com/docs/reference/context-usage/.
Step 3: Read specific files (if detailed question)
Follow the lookup rule. Start at skills/cf-help/topics.md, then skills/<name>/SKILL.md, agents/<name>.md, hooks/. Config keys and custom guides: topics.md. Repo-only schema: docs/config-schema.md.
Step 4: Common Workflows (if workflow question)
Present the workflows in topics.md. Distinction: $cf-remember = project knowledge for AI; $cf-learn = notes for the human.
Step 5: Troubleshooting (if troubleshooting question)
- Skill not triggering? Invoke
/cf-<skill-name> manually.
- Config / custom guides?
topics.md.
- Permission prompts? Native modes: Claude
auto / acceptEdits / sandbox.autoAllowBashIfSandboxed; Codex approval_policy / Smart Approvals / --approve-for-me; AGY remembers per-conversation; omp yolo. CF hook: autoApprove. Details: topics.md.
- More? Troubleshooting.
Step 6: Answer concisely
Answer from sources you read. Link files for deeper dives. If you skipped the lookup rule on a flag / hook / CLI / config question, go back.
CLI Requirements (quick reference)
Plugin works without CLI. NONE / OPTIONAL (MCP or grep) / REQUIRED (none today). Matrix: docs/cli-requirements.md.
1---2name: cf-help-23description: Answer questions about Coding Friend itself — skills, agents, hooks, CLI, config, memory, hosts, parameters. TRIGGER — "what skills are available?", "how does coding friend work?", "list all skills", "how do I use cf-plan?", "what is cf-tdd?", "is the CLI required?", or any question about a cf-* name, flag, hook, or config key. SKIP — general coding questions unrelated to Coding Friend, and requests to run a skill (invoke that skill instead).4---56# $cf-help — Coding Friend Help78Catalog = what exists. Flags / config / hooks → lookup rule. Never guess.910## Workflow1112### Step 0: Custom Guide1314```!15bash "${PLUGIN_ROOT}/lib/load-custom-guide.sh" cf-help16```1718If the block above printed anything, apply only the `## Before`, `## Rules`, and `## After` sections; if it shows the raw command instead of output, re-run that exact `load-custom-guide.sh` fence now.1920### Step 1: Understand the question2122Classify (more than one is fine): general overview · specific skill · specific agent · hooks / fewer prompts · CLI · setup / config · memory / MCP · workflow · troubleshooting.2324### Lookup rule (do not skip)2526Never invent a flag, config key, hook name, CLI command, or host difference.27281. "What exists / which skill?" → catalog below.292. Anything more specific → Read `${PLUGIN_ROOT}/skills/cf-help/topics.md` (or `plugin/skills/cf-help/topics.md` in this repo) — topic → source map, flags, hooks, CLI, config, native prompt reduction.303. Then Read the source that index points to (`skills/<name>/SKILL.md`, `agents/<name>.md`, `hooks/<file>`).314. If still unclear: Glob / Grep under `${PLUGIN_ROOT}/skills/`, `agents/`, `hooks/`.3233### Step 2: Provide overview (if general question)3435Coding Friend is a lean toolkit for disciplined engineering workflows in Codex CLI. Check skills first; TDD optional (`--add-tests` or `tdd: true`); verify before claiming done.3637Hosts: Claude Code (default), Codex CLI, omp, **Google Antigravity** (`--agent agy` / `--agy`). Plugin works alone; `coding-friend-cli` is optional. Skills never call `cf`. Flags / config / native prompt-reduction: `topics.md`.3839### Slash Commands (user triggers with /)4041- `$cf-advise [decision]` — ⚡⚡ — Advisory interview. `--quick`, `--save`42- `$cf-ask [question]` — ⚡⚡ — Codebase Q&A → docs/memory/43- `$cf-plan [task]` — ⚡⚡ — Phased plans. `--fast`/`--quick`, `--hard`, `--auto`, `--inline`/`--no-file`, `--gui`/`--human`, `--model <name>` pin the model for cf-planner at the brainstorm step (Codex model name, e.g. `gpt-5.5`; Claude aliases are not valid). Autopilot review-fix cap: `review.maxRounds` (default 5). Normal/hard mode also writes `brief.md` (request, Q&A, assumptions) for `$cf-plan-review`.44- `$cf-plan-resume <plan>` — ⚡⚡ — Resume a saved plan. Honors `auto: true`.45- `$cf-plan-review [plan]` — ⚡⚡ — Review a saved plan with a fresh in-session reviewer before implementing. `--codex`, `--gemini`, `--claude`, `--cursor`, `--grok` add external reviewers in parallel (the flag matching the current host is skipped). Writes `review.md` into the plan folder, offers to apply Critical/Important findings.46- `$cf-later-do [item]` — ⚡⚡ — Resolve `docs/later/` via `$cf-fix` or `$cf-plan`47- `$cf-review [target]` — ⚡⚡ — Dispatch review. Flags: `--claude`, `--gemini`, `--cursor`, `--grok` run headless external reviewers in parallel and merge into one report; `--out` exports a `$cf-review-out` prompt with in-session findings embedded. `--with-codex`/`--codex` and `review.withCodex` are ignored on Codex (do not spawn a nested Codex review). `review.agentTimeout` (default 300s) bounds each external agent; `review.maxRounds` (default 5) caps the autopilot fix loop. Unavailable agents are skipped with a warning.48- `$cf-review-out [label]` — ⚡⚡ — Prompt + diff → `docs/reviews/`49- `$cf-review-in <label> [service]` — ⚡⚡ — Read external review, offer to fix50- `$cf-commit [hint]` — ⚡ — Conventional commit51- `$cf-design [mode]` — ⚡⚡ — Scan / design / modify UI52- `$cf-ship [hint]` — ⚡ — Verify, commit, push, PR (`--dry-run`)53- `$cf-fix [bug]` — ⚡⚡ — Quick bug-fix54- `$cf-optimize [target]` — ⚡⚡ — Baseline → optimize → measure55- `$cf-scan [description]` — ⚡⚡⚡ — Bootstrap memory56- `$cf-remember [topic]` — ⚡⚡ — Project knowledge → docs/memory/. Also auto-invoked.57- `$cf-learn [topic]` — ⚡⚡ — Educational notes. Also auto-invoked.58- `$cf-teach [topic]` — ⚡⚡ — Story of what happened59- `$cf-research [topic]` — ⚡⚡ — Web research → docs/research/60- `$cf-session [label]` — ⚡⚡ — Cross-machine session save61- `$cf-warm [user]` — ⚡⚡ — Git history catch-up62- `$cf-checkpoint [additional-prompt]` — ⚡⚡ — Checkpoint → docs/context/checkpoints/63- `$cf-checkpoint-from <slug> [message]` — ⚡⚡ — Restore checkpoint, then act. `--recap`64- `$cf-help [question]` — ⚡⚡⚡ — This skill. Also auto-invoked.6566### Auto-Invoked Skills (activate automatically when relevant)6768- **cf-tdd** — ⚡⚡ — Code-writing gate. Default direct; TDD with `--add-tests` or `tdd: true`. `--auto` = review + fix + commit (`review.maxRounds`, default 5).69- **cf-sys-debug** — ⚡⚡ — 4-phase debug70- **cf-verification** — ⚡ — Evidence gate before claiming done71- **cf-learn** / **cf-remember** / **cf-help** — also auto-invoked (see above)7273### Agents (run in forked sessions — separate context window)7475- **cf-reviewer** — ⚡ — Orchestrator: **cf-reviewer-plan** (medium reasoning effort), **cf-reviewer-security** (medium reasoning effort), **cf-reviewer-quality** (low reasoning effort), **cf-reviewer-tests** (low reasoning effort), **cf-reviewer-rules** (low reasoning effort), **cf-reviewer-reducer** (low reasoning effort)76- **cf-implementer** — ⚡ — Writes code; TDD with `--add-tests`. `[CF-RESULT: success|failure]`. No autopilot.77- **cf-explorer** — ⚡ — Repo map + context files78- **cf-planner** — ⚡ — Approaches + phased tasks79- **cf-writer** — ⚡ — Straightforward markdown80- **cf-writer-deep** — ⚡ — Nuanced technical docs8182### Hooks (automatic — not slash commands)8384`hooks/hooks.json` (+ `*.agy.*`; Codex transformed). Keys + native modes: `topics.md`.8586**session-init.sh** · **rules-reminder.sh** · **privacy-block.sh** · **scout-block.cjs** · **auto-approve.cjs** (`autoApprove`; Claude `autoApproveLLM` default false → unknown defers to native) · **session-log.sh** · **task-tracker.sh** · **agent-tracker.sh** · **memory-capture.sh** · **statusline.sh** (Claude; `cf statusline`)8788### CLI (`coding-friend-cli`, binary `cf`) — optional8990Lifecycle `install|uninstall|enable|disable|update` (`--user|--project|--local`, `--agent claude|codex|omp|agy`); setup `init` `config` `permission` `statusline`; memory / learn / mcp; `status` `clean` `session` `guide` `dev`. No skill requires the CLI. Full flags: `topics.md`.9192Tiers `⚡` / `⚡⚡` / `⚡⚡⚡`: https://cf.dinhanhthi.com/docs/reference/context-usage/.9394### Step 3: Read specific files (if detailed question)9596Follow the lookup rule. Start at `skills/cf-help/topics.md`, then `skills/<name>/SKILL.md`, `agents/<name>.md`, `hooks/`. Config keys and custom guides: `topics.md`. Repo-only schema: `docs/config-schema.md`.9798### Step 4: Common Workflows (if workflow question)99100Present the workflows in `topics.md`. Distinction: `$cf-remember` = project knowledge for AI; `$cf-learn` = notes for the human.101102### Step 5: Troubleshooting (if troubleshooting question)103104- **Skill not triggering?** Invoke `/cf-<skill-name>` manually.105- **Config / custom guides?** `topics.md`.106- **Permission prompts?** Native modes: Claude `auto` / `acceptEdits` / `sandbox.autoAllowBashIfSandboxed`; Codex `approval_policy` / Smart Approvals / `--approve-for-me`; AGY remembers per-conversation; omp `yolo`. CF hook: `autoApprove`. Details: `topics.md`.107- **More?** [Troubleshooting](https://cf.dinhanhthi.com/docs/reference/troubleshooting/).108109### Step 6: Answer concisely110111Answer from sources you read. Link files for deeper dives. If you skipped the lookup rule on a flag / hook / CLI / config question, go back.112113## CLI Requirements (quick reference)114115Plugin works without CLI. **NONE** / **OPTIONAL** (MCP or grep) / **REQUIRED** (none today). Matrix: [docs/cli-requirements.md](../../../docs/cli-requirements.md).
dinhanhthi