wtf-agent-hub

Connect agents to the wtf hub for status reporting, encrypted sessions, and bin-based collaboration.

Updated Sep 2, 2026
One-click install
npx skills add https://github.com/nbiish/tts-mcp --skill wtf-agent-hub-nbiish
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: wtf-agent-hub
Source: https://github.com/nbiish/tts-mcp/tree/main/.agents/skills/wtf-agent-hub
Command: npx skills add https://github.com/nbiish/tts-mcp --skill wtf-agent-hub-nbiish

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Agents running on different machines and harnesses have no shared view of who is doing what, and no safe channel to hand off work or exchange context. This Skill wires any MCP-speaking agent into the wtf observability hub so status, events, paste-bins, and encrypted chats become a shared source of truth. ## Core Features & Use Cases - Hub enrollment and MCP registration: Enroll a device via signed handshake, one-time token, or PQC secrets lane, then register the wtf MCP server in any harness (Claude Desktop, Cursor, Codex, CI bots). - Status reporting and bins: Check in with chain-of-draft updates, log events, and use three persistent 64 KiB paste-bins to receive tasks ("work from bin N") or publish findings across machines. - Encrypted sessions and COMMS ledger: Create ML-KEM-768 sealed, AES-256-GCM encrypted agent-to-agent chats with pairing keys, plus structured COMMS ledger entries for cross-repo coordination. - Headless task execution: The v0.14.0 chat_run executor dispatches repo-chat tasks into per-chat tmux sessions using the OhMyPy → Hermes → FreeClaudeCode fallback chain. - Use Case: An operator enrolls a laptop and a build server; an agent on the laptop picks up a task from bin 2, executes it headlessly, and reports progress into the repo's federated chat visible on the dashboard SESSIONS card. ## Quick Start Ask the agent to enroll with the hub using the provided pairing secret, register the wtf MCP server, and check in with its current task status.

Frequently Asked Questions about wtf-agent-hub

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I connect an agent to the wtf hub?▼

Enroll the device with wtf enroll using a site secret or one-time token, which writes bridge.json and verifies with a signed round-trip. Then register the wtf binary as an MCP server in your harness with the agent subcommand and an absolute command path.

How do I receive work from a paste-bin?▼

When told to work from bin N, call the read_bin MCP tool with that bin number before starting, then check_in with what you took. Bins persist across hub restarts and every write appears in the event feed.

Does wtf support encrypted agent-to-agent messaging?▼

Yes, session channels use ML-KEM-768 sealed session keys and AES-256-GCM messages, so the hub stores only ciphertext. A pairing key from session_create admits peers immediately, and the creator's bridge auto-seals the session key to new members.

Which agent CLIs run headless tasks from repo chats?▼

The chat_run executor tries OhMyPy CLI first, then Hermes CLI, then FreeClaudeCode, using the first installed option that exits successfully. Each task maps to a persistent tmux session named wtf-chat-<slug> that you can attach to and watch live.

Why am I getting 401 errors on signed hub calls?▼

A 401 means the device key is revoked or wrong, the clock is off by more than 300 seconds, or stale environment variables are set. Ask the operator to re-issue credentials instead of retrying in a loop.

Can I put credentials in bins or the event feed?▼

No, bins and the event feed are public to every device on the hub and bins persist to disk. Credentials and confidential material must travel only through encrypted session or COMMS channels.