> ## Documentation Index
> Fetch the complete documentation index at: https://foreverbetter.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# Agent skill

> Give an agent one executable instruction file for authentication, discovery, data connection, analysis, and delivery.

The hosted `SKILL.md` is the agent entry point for ForeverBetter. It tells an
agent how to choose cloud, self-hosted, or local execution; authenticate without
putting credentials in chat; discover the live deployment; and complete a health
workflow instead of stopping at setup instructions.

<CardGroup cols={2}>
  <Card title="Open the skill file" icon="file-code" href="https://api.foreverbetter.xyz/SKILL.md">
    Read or download the canonical, machine-readable instruction file.
  </Card>

  <Card title="Connect your agent" icon="link" href="/connect-your-agent">
    Follow the browser handoff and attach the API over MCP.
  </Card>
</CardGroup>

## Give it to an agent

Paste this instruction into any agent that can read URLs:

```text theme={null}
Help me analyze and connect my longevity data.
Read https://api.foreverbetter.xyz/SKILL.md and execute its onboarding workflow.
Do the safe steps yourself, ask me only for required choices or browser approvals,
and finish by delivering the result I choose.
```

Agents can also fetch it directly:

```bash theme={null}
curl -fsSL https://api.foreverbetter.xyz/SKILL.md
```

## What makes the workflow agentic

The skill gives the agent an operating contract, not only API documentation:

1. Choose cloud, self-hosted API, or the local file pipeline.
2. Use browser-based agent login so the user never pastes an OTP, API key, OAuth
   code, or OAuth state into chat.
3. Read the live manifest, capabilities registry, endpoint catalog, and OpenAPI
   contract before acting.
4. Ask which outcome the user wants and which data is already available. A
   dashboard is optional: the result can instead be a one-off or recurring action
   plan, ancestry result, trend loop, health agent, or full multimodal optimization.
5. Connect the minimum useful data, then run the matching playbook end to end.
6. Poll queued jobs and wearable connections until they complete or fail.
7. Deliver the dashboard, action plan, ancestry result, provider shortlist, or
   source-backed answer with coverage and provenance.

The agent asks before a paid call, destructive deletion, external share, or
public upload. It does not repeatedly ask for confirmation for ordinary reads,
analysis, or creation of private artifacts after the user chooses a workflow.
Cloud users can use a subscription or x402 per-call payment on supported routes;
self-hosted and local-pipeline modes do not require a hosted subscription.

## Live discovery wins

The skill deliberately avoids freezing deployment details into an agent prompt.
An agent checks these public contracts before it connects data:

| Contract                             | Purpose                                                             |
| ------------------------------------ | ------------------------------------------------------------------- |
| `GET /.well-known/health-agent.json` | Authentication, MCP tools, scopes, and agent-login sequence         |
| `GET /capabilities`                  | Available modalities, providers, connection types, and endpoint IDs |
| `GET /endpoints`                     | Method, path, scope, and grant catalog                              |
| `GET /openapi.json`                  | Request and response schemas                                        |
| `/llms-full.txt`                     | Complete human documentation as agent-readable text                 |

If a static example disagrees with these live contracts, the live deployment is
authoritative.

## Completion standard

Setup is not the result. A run is complete only when the agent delivers the
outcome the user selected or clearly reports the terminal blocker and preserves
partial progress. Missing modalities appear as optional coverage gaps; one usable
source is enough to produce a first result.

For the lower-level MCP transport and tool-call format, continue to
[Agents and MCP](/agents-and-mcp).