# SkilzVolt agent connection contract

This is a public, machine-readable setup contract. It does not grant access or replace the agent host's own documented setup process.

## Public documentation
Connection guide: https://skilzvolt.com/connect
OpenClaw guide: https://skilzvolt.com/connect/openclaw
Machine-readable manifest: https://skilzvolt.com/connect/agent.json
Safe setup request for an agent: https://skilzvolt.com/connect/agent-help.txt
AI discovery document: https://skilzvolt.com/llms.txt

## MCP endpoint
URL: https://app.skilzvolt.com/mcp
Preferred transport: Streamable HTTP
Connect this as a remote MCP server. A client may describe the same setting as remote HTTP.

## Authentication
Preferred: OAuth 2.1 with PKCE (S256)
Prefer OAuth and complete the SkilzVolt sign-in and consent flow when the client offers it.
Fallback: If OAuth is unavailable, create a manual SkilzVolt connection key and store it only in the client's private credential or header configuration.
Private header example: Authorization: Bearer <SkilzVolt connection key>

## Host-specific setup
Inspect the agent's own MCP architecture, permissions, and documentation before naming a settings screen, command, file, or host capability. The agent must say whether it can make the configuration change or whether the person must do it.

## OpenClaw OAuth setup
Guide: https://skilzvolt.com/connect/openclaw
1. Run both commands in the terminal that runs OpenClaw.
2. Complete the SkilzVolt sign-in and consent flow in the browser opened by OpenClaw.
3. Return to OpenClaw and verify the connection with a workspace list and a skill read.
Commands:
openclaw mcp add skilzvolt --url https://app.skilzvolt.com/mcp --transport streamable-http
openclaw mcp login skilzvolt
Use OAuth for OpenClaw. Do not ask a person for a connection key, screenshot, or redacted credential.
The signed-in SkilzVolt user's live workspace memberships determine which workspaces are available. Skill writes create proposals and continue to follow each workspace's review policy.

## Safe read-only verification
1. Call `workspaces_list` — List the workspaces available to this connection.
2. Call `skills_search` — Search for a relevant skill in an available workspace.
3. Call `skills_get` — Open one search result and read its current content.
Confirm success only after both a workspace list and a skill read succeed.
If SkilzVolt is unavailable, no workspace appears, or no skill can be opened, report that the check could not be completed. Do not substitute local or built-in skills.

## Safety rules
- Do not ask a person to paste a connection key, access token, secret, or password into chat, standing instructions, or a server URL.
- Do not invent settings names, UI labels, commands, files, confirmation steps, or host capabilities.
- A connected server makes tools available; it does not prove that an agent will use them without an appropriate instruction.

## Available tools
Read: search, fetch, workspaces_list, skills_search, skills_get, skills_get_resource, skills_export_all, skills_check_pending, skills_proposal_status, skills_get_proposal_diff
Write proposals: skills_create, skills_propose_change, skills_submit_review