multitable

Standing pointer to the external `omnigent-ai/omnigent` project — MultiTable's closest prior-art / sibling. Trigger when the user mentions omnigent, omni, omnigent-ai, "compare to omnigent", "how does omnigent do X", `claude_native.py` / `codex_native.py` / `cursor_native.py` / `pi_native.py`, omnigent's `runner/` / `server/` / `policies/`, omnigent policy engine, omnigent governance, omnigent host registry / presence / attach / co-drive, omnigent cloud sandbox (Modal / Daytona / Islo), or wants a side-by-side of an omnigent concept against MultiTable.

Standing pointer to the broader landscape of open-source agent harnesses, meta-harnesses, governance layers, multi-agent orchestrators, and usage/cost observability sidecars that solve problems adjacent to MultiTable. **Consult proactively** when (a) the user explicitly names a project — Mastra, OpenHarness, "Microsoft Agent Governance Toolkit", Cordum, CrewAI, LangGraph, "Strands Agents", "Harness SDK", VoltAgent, MetaGPT, ruflo, "Claude Flow", artificial (the Go harness), "Claude Squad", OpenCode (sst/opencode), Archon, CodexBar, awesome-agent-harness, awesome-harness-engineering, "9 open-source agent orchestrators" — or asks "is there an OSS project that does X", "what's prior art for Y", "similar projects to MultiTable", "harness comparison", "agent governance frameworks", "meta-harness", "multi-agent orchestrator", "cross-provider usage limits", "menu-bar usage tracker"; **AND ALSO** (b) work is stuck on a persistent bug that's resisted two or more fix attempts in an area covered by the parallels table (

Authoritative reference for working on MultiTable's Claude Agent SDK integration (`@anthropic-ai/claude-agent-sdk`). Trigger when the user mentions the agent SDK, `query()`, streaming, permission modes, `canUseTool`, `AskUserQuestion`, elicitation, hooks, abort/stop, plan mode, sessions, subagents, MCP, usage limits / rate limits (`rate_limit_event`, `rate_limit_info`), or modifying anything under `packages/daemon/src/agent/` or `packages/daemon/src/hooks/`.

Authoritative reference for working on MultiTable's Cursor provider, driven by the Cursor CLI (`cursor-agent`) in headless `--print --output-format stream-json` mode. Trigger when the user mentions Cursor, cursor-agent, the Cursor adapter, CursorAdapter, stream-json / NDJSON events, `--resume` / `--continue`, `--mode plan|ask`, `--force` / `--yolo`, `--sandbox`, composer-2.5, `cursor-agent models`, `~/.cursor`, agent-transcripts, usage limits / rate limits (why they're off for Cursor; the cursor.com billing out-of-band path), or modifying anything under `packages/daemon/src/agent/providers/cursor.ts`, `packages/daemon/src/agent/providers/cursor-cli/`, or `packages/daemon/src/transcripts/cursorParser.ts`.

Authoritative reference for working on MultiTable's GitHub Copilot SDK integration (`@github/copilot-sdk`). Trigger when the user mentions the Copilot SDK, `CopilotClient`, `CopilotSession`, `session.send`, `sendAndWait`, `session.idle`, `assistant.message_delta`, `onPermissionRequest`, `onUserInputRequest`, `onElicitationRequest`, plan / autopilot / chat mode for Copilot, session.abort, `~/.copilot/session-state`, or adding/modifying a Copilot provider under `packages/daemon/src/agent/providers/`.

Authoritative reference for working on MultiTable's Grok Build (xAI) provider, driven by `grok agent stdio` over line-delimited JSON-RPC (Agent Client Protocol). Trigger when the user mentions Grok Build, the Grok adapter, `grok agent stdio`, GrokAdapter, grok-build-0.1, ACP for Grok, `~/.grok/`, the default/auto/plan modes, `--always-approve`, `--agent-profile`, `--reasoning-effort`, `x.ai/billing`, the SuperGrok/SuperHeavy subscription, `GROK_CODE_XAI_API_KEY`, usage limits / rate limits (why they're off for Grok; the xAI billing out-of-band path), or modifying anything under `packages/daemon/src/agent/providers/grok.ts`, `packages/daemon/src/agent/providers/grok-acp/`, or `packages/daemon/src/transcripts/grokParser.ts`.

Authoritative reference for working on MultiTable's Hermes (xAI Grok) provider, driven by `hermes acp` over line-delimited JSON-RPC (Agent Client Protocol). Trigger when the user mentions Hermes, the Hermes adapter, ACP, `hermes acp`, Grok / grok-4.3, xAI OAuth, `session/prompt`, `session/update`, `session/request_permission`, `~/.hermes/`, the `/reasoning` slash command, usage limits / rate limits (why they're off for Hermes; the xAI billing out-of-band path), or modifying anything under `packages/daemon/src/agent/providers/hermes.ts`, `packages/daemon/src/agent/providers/hermes-acp/`, or `packages/daemon/src/transcripts/hermesParser.ts`.

Authoritative reference for working on MultiTable's OpenAI Codex SDK integration (`@openai/codex-sdk`). Trigger when the user mentions the Codex SDK, `Codex`, `Thread`, `runStreamed`, `ThreadEvent`, `ThreadItem`, sandbox/approval policy, `~/.codex/sessions`, `resumeThread`, `codex exec`, plan-mode for Codex, usage limits / rate limits (`account/rateLimits/updated`, `RateLimitSnapshot`), or modifying anything under `packages/daemon/src/agent/providers/codex.ts` or `packages/daemon/src/transcripts/codexParser.ts`.

Behavioral guidelines to reduce common LLM coding mistakes. Use when writing, reviewing, or refactoring code to avoid overcomplication, make surgical changes, surface assumptions, and define verifiable success criteria.

Build web-based terminals by wiring node-pty on a Node.js backend to xterm.js in a React frontend over WebSockets. Use this skill whenever the user mentions node-pty, wants to embed a shell/terminal in a web UI, build a browser-based SSH or REPL interface, a cloud IDE terminal, a remote shell tool, a dev-container console, or anything that connects xterm.js to a real PTY — even if they don't name the library. Also use for Express + ws terminal servers, ConPTY on Windows, PTY resize/fit handling, multi-session terminal backends, and reconnection/session persistence for web terminals.