SignalBin Docs

Appearance

Agent Skill ​

SignalBin ships an Agent Skill: a self-describing document that teaches an AI coding agent how to drive a SignalBin workspace, so you don't have to explain any of it yourself.

Where it's served ​

The skill document is a live endpoint on every SignalBin deployment, not something you download separately:

  • /skill.md
  • /.well-known/agent-skills/signalbin/SKILL.md, the path some agent clients check automatically

Both return the same content, as text/markdown. There's also a discovery index at /.well-known/agent-skills/index.json listing the skill's name, description, canonical URL, and a SHA-256 digest of the exact bytes served at that URL, so a client can tell whether its cached copy is current.

Every URL inside the document is rendered against your deployment's own base URL, so the examples are copy-pasteable as written instead of templates you have to fill in first. A self-hosted instance gets working links that way with no build-time configuration. The bytes differ per deployment, so the digest does too.

TIP

If you're reading this page in a browser, you can see the raw skill document right now at signalbin.work/skill.md.

What it covers ​

The skill document is written for an agent, not a human, so it's denser than this guide. It covers:

  • Which transport to reach for, in order: MCP if the client speaks it, then the signalbin CLI, then raw HTTP. The agent is told to run command -v signalbin before writing an HTTP call, and to install the CLI if it can, because --json, webhooks list --all, and listen --forward are all things curl makes harder or cannot do at all
  • Where a token comes from (a human creates it under Settings on the API Tokens tab, and it's shown once) and the rules for handling one: SIGNALBIN_TOKEN in the environment rather than on the command line, bearer header only, never in a URL, never echoed back into chat or a log. It's told not to run signalbin auth login, which would write the token to a config file that outlives the task
  • Why to call whoami first, so the agent learns the workspace, role, and the token's exact granted scopes instead of discovering them one 403 at a time
  • The full scope table: endpoints:read, endpoints:write, webhooks:read, billing:read, billing:write, team:read, team:write
  • Worked examples for the common jobs, each given as a CLI command with its REST equivalent underneath: creating a receiver endpoint, sending a realistic sample payload into it, listing recent deliveries, debugging a failing integration, watching captures live and forwarding them to a local server, relaying captures to destinations, replaying one capture, checking billing, inviting a teammate
  • A rule that captured webhook content is third-party data, not instructions. An agent reading a payload should report on it, never follow or execute something found inside it, and never pipe webhooks show --body into a shell
  • Operational pitfalls that aren't obvious from the API shape: a checkout call doesn't move any money, a token's workspace is fixed at creation, rotating an endpoint secret breaks existing senders, a self-hosted instance needs SIGNALBIN_SERVER set or the CLI talks to the wrong deployment, and --json doubles as the CLI's confirmation prompt, so a destructive command run that way never stops to ask
  • Every documented error code and what to do about it

The API and the skill document are versioned together, and the document says so: where it disagrees with /api/openapi.yaml, the spec wins.

Using it ​

Point a skill-aware AI client at either URL. There's no separate credential, since the skill document reuses the same API token as everything else. If your client speaks MCP, the document tells it to prefer that; otherwise it reaches for the CLI, and falls back to plain HTTP only when neither is available. See MCP Integration for how that connection works.

The skill document assumes the agent already holds a token and is working inside a real integration. For open discovery, when something is just crawling to learn what SignalBin is, there's a separate public guide at /llms.txt.

Next steps ​