Skip to content
Back to skills

Help

ASecurity

List available commands the user can invoke. Works from CLI or messaging channels. Triggers on /help, /commands, /agent:help, "qué comandos puedo usar", "help commands".

  • 63 stars
  • 0 votes
  • 0 copies
  • 1 view
  • Added May 26, 2026
ai-agents

Works with

  • claude code
  • cli
  • mcp

Security analysis

A100/100

Scanned May 27, 2026

npx -y skills add crisandrews/ClawCode --skill help --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Help?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Help
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/crisandrews-help/badge)](https://www.skillsdirectory.com/skills/crisandrews-help)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: help
description: List available commands the user can invoke. Works from CLI or messaging channels. Triggers on /help, /commands, /agent:help, "qué comandos puedo usar", "help commands".
user-invocable: true
---

# Help — list commands

Show a concise, LIVE list of all available commands. Never hardcode — always pull from the `list_commands` MCP tool so recently installed or removed skills are reflected.

## Steps

1. **Detect the surface.** If the incoming message has a `<channel source="...">` marker, you're on a messaging channel. Otherwise you're on CLI or WebChat.

2. **Call `list_commands`** with the right format:
   - **CLI / WebChat:** `list_commands({ format: "table", includeTools: false })` — grouped markdown, readable.
   - **Messaging (WhatsApp, Telegram, Discord, iMessage):** `list_commands({ format: "compact", includeTools: false })` — one line per command, mobile-friendly.

   Leave `includeTools: false` for the user-facing help — MCP tools are agent-invoked, not user-typed. Users who want the internals can ask "what MCP tools do you have?" and you call with `includeTools: true`.

3. **Print the output** from the tool as-is. Do not rewrite or reorder.

4. **Adapt formatting to the channel** after the tool output:
   - WhatsApp: strip `**bold**` → `*bold*`, remove markdown headers if too noisy
   - Telegram: keep `**bold**` or convert to HTML
   - Discord: markdown as-is
   - iMessage: strip markdown to plain text
   - CLI / WebChat: markdown as-is

5. **Append a short native-tools note** at the end (only on CLI):
   ```
   Native Claude Code commands (CLI only): /status /usage /cost /compact /clear /mcp /model /help
   ```

## Example flows

### CLI user asks `/help`

Call `list_commands({ format: "table", includeTools: false })`. Print the returned markdown. Append the native-tools note.

### WhatsApp user says "what can you do?"

Call `list_commands({ format: "compact", includeTools: false })`. Convert any `**bold**` to `*bold*`. Don't include the native-tools note (doesn't apply).

## Notes

- `/help` and `/commands` are aliases.
- Keep the response short on mobile channels. `compact` format handles that.
- If the user wants the internals ("tools", "MCP"), call with `includeTools: true` so they see them too.
- If nothing comes back, something's wrong with the install — suggest `/agent:doctor`.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.