Add provider CLI commands and skills discovery

[matheus1lva]

What Changed

Added filesystem discovery for CLI slash commands and skills from Codex and Claude providers. The implementation includes:

1. New discovery module (providerCommandsDiscovery.ts): Scans user and project directories for markdown-based commands and skills, extracting metadata from frontmatter or content.

2. New contracts (providerCommands.ts): Schemas for ProviderCommandEntry, ProviderCommandsListInput/Result, and error types.

3. RPC integration: Added providers.listCommands endpoint to the WebSocket RPC layer and native API.

4. Web UI updates:

   • New React Query hook (providerCommandsReactQuery.ts) for fetching discovered commands/skills
   • Extended composer trigger detection to recognize $skill syntax alongside /command
   • Updated ComposerCommandMenu to display CLI commands and skills with appropriate icons
   • Integrated discovered commands/skills into the composer autocomplete menu

5. Composer logic: Extended to detect and handle both /command (CLI commands) and $skill (CLI skills) triggers, with proper filtering and insertion.

Why

This enables the web composer to surface CLI commands and skills that users have defined locally (in ~/.codex/commands, ~/.claude/commands, etc.) or at the project level (./.codex/commands, ./.claude/commands, etc.), providing a seamless autocomplete experience similar to the native CLI tools. Project-scoped commands take precedence over user-scoped ones with the same name.

UI Changes

The composer now shows:

• Built-in slash commands (/model, /plan, /default) with a bot icon
• Discovered CLI commands with a terminal icon
• Discovered CLI skills with a sparkles icon
• Descriptions extracted from markdown frontmatter or first non-empty line
• Source badges indicating whether commands are from user or project scope

Checklist

• This PR is small and focused
• I explained what changed and why
• Existing tests pass (added unit tests for trigger detection in composer-logic.test.ts)

https://claude.ai/code/session_018bAJtpkRpiSfFmnmdASfvV

---

Note

Medium Risk
Adds a new filesystem-scanning RPC that reads user/project directories and wires results into the chat composer autocomplete; risk is moderate due to new IO paths and query-triggered UI behavior, but scope is contained and errors are wrapped.

Overview
Adds end-to-end discovery and autocomplete for provider CLI commands and skills.

The server now scans user and project .~provider directories (plus Codex prompts) for markdown-defined commands/skills, extracts descriptions, dedupes by name with project overriding user, and exposes results via a new providers.listCommands WebSocket RPC with new providerCommands contract schemas.

The web app queries this RPC (React Query) and extends the composer to suggest and insert discovered /${command} items alongside built-ins, and adds $skill trigger detection and menu rendering (new item types, icons, filtering, and insertion behavior), with updated unit tests for trigger parsing.

^{Reviewed by Cursor Bugbot for commit 48f1b58. Bugbot is set up for automated code reviews on this repo. Configure here.}

Note

Add provider CLI commands and skills discovery to the composer menu

• Adds a discoverProviderCommands Effect in providerCommandsDiscovery.ts that scans ~/.<provider>/ and <cwd>/.<provider>/ directories for .md command files and skill folders, deduplicating by source priority (project > user > builtin).
• Wires discovery through a new providers.listCommands RPC (WebSocket method, contracts, server handler, and NativeApi passthrough) so the web layer can fetch results via React Query.
• Extends detectComposerTrigger so any /<token> opens the slash-command menu (except /model) and $<token> opens a new cli-skill menu; the composer fetches and displays discovered commands and skills accordingly.
• Behavioral Change: /-prefixed tokens no longer require whitelisting — any unknown slash token now triggers the command menu instead of being ignored.

^{Macroscope summarized 48f1b58.}

[coderabbitai]

Important

Review skipped

Auto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the .coderabbit.yaml file in this repository. To trigger a single review, invoke the @coderabbitai review command.

⚙️ Run configuration

Configuration used: Repository UI

Review profile: CHILL

Plan: Pro

Run ID: d995f487-00a3-45f8-aa6b-7de691af60c3

You can disable this status message by setting the reviews.review_status to false in the CodeRabbit configuration file.

Use the checkbox below for a quick retry:

• 🔍 Trigger review

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 48f1b58. Configure here.}

[cursor]

Loading state ignores provider commands fetch status

Low Severity

isComposerMenuLoading only considers the "path" trigger kind, completely ignoring the fetching state of providerCommandsQuery. When a $skill trigger is activated for the first time (no cached data), the menu displays "No matching skill." instead of "Searching..." while the query is in-flight. This is especially noticeable for the cli-skill trigger since it has no built-in fallback items to fill the menu during loading.

Additional Locations (1)

• apps/web/src/components/ChatView.tsx#L1448-L1459

 

^{Reviewed by Cursor Bugbot for commit 48f1b58. Configure here.}

[macroscopeapp]

Approvability

Verdict: Needs human review

This PR introduces a new feature for discovering and autocompleting provider CLI commands and skills from the filesystem. It adds new server-side discovery logic, API contracts, and UI integrations across multiple packages - the scope of new capabilities warrants human review.

^{You can customize Macroscope's approvability policy. Learn more.}