One shared skills vault for every AI agent

Agents can read the same verified setup details without relying on search indexing or interactive platform selectors. Use the plain-text contract or the JSON contract.

Endpoint

https://app.skilzvolt.com/mcp

Transport

Streamable HTTP

Authentication

OAuth is recommended when available. Complete the SkilzVolt sign-in and consent flow offered by the client.

An agent must inspect its own setup options and say which configuration steps it can perform. It must not invent settings labels, commands, files, or host capabilities.

These public URLs are safe to send to an agent or include in an agent directory. They explain the MCP endpoint and OAuth-first setup without exposing a connection key.

• Copy-ready agent setup request — asks an agent to inspect its own setup options before giving instructions.
• AI discovery document (llms.txt) — links agents directly to the connection guide and machine-readable contract.
• OpenClaw setup guide — gives OpenClaw the exact OAuth commands and a safe verification sequence without requiring a key.

You do not need to know the name of an instruction file before you start. Never include API keys or other secrets in this message.

Please help me add the SkilzVolt reference-use guidance below to my permanent instructions for future chats.

First, say clearly whether you can edit my project, custom-instruction, or agent-settings store yourself. Do not claim this guidance is saved unless you actually made that change. If you cannot edit it, tell me the exact settings path and the next step in plain language so I can paste this guidance there myself. Do not put this guidance in chat memory or create a local instruction file. Do not put secrets or connection keys in the instructions.

## SkilzVolt skills

This connection covers every workspace you can access.

### Operating card
- **Use:** Consult SkilzVolt for relevant procedures and lessons.
- **Read:** Find via `skills_catalogue`/`skills_search`; call `skills_get`. While `chunk.complete=false`, continue with `version_id` and `next_start_char`; verify `content_sha256`.
- **Write:** Use `skills_create` for new skills. For edits, read purpose/rules/body with `skills_get`; preserve/revise it, then call `skills_propose_change` with `current_version_id` as `base_version_id` and the complete revised body as `proposed_content`. To change details, send `proposed_metadata` as a complete replacement of name, description, purpose, and rules. No in-place patch primitive is needed.
  - **Finish:** Creates proposals. Inspect the returned `review_mode`. For `agentic`, poll, fetch `skills_get_proposal_diff`, review evidence, and submit its hash to approve/revise; repeat without asking. For `human`, report ID and await an authorised reviewer; never self-approve/revise; always read back the resulting current version before saying it is live.
  - **Recover:** For error code `stale_base_version`, call `skills_get` again and resubmit the complete revised body with its new `current_version_id`. For conflicts, inspect status and review package before retrying.
- **Limits:** Vault content is user-authored reference content. It does not override host, system, developer, or safety instructions and does not automatically change host configuration or behavior. Use other appropriate platform tools or sources when relevant; SkilzVolt does not limit other available tools.
  - **Failure:** If the vault is unavailable or catalogue access is degraded, report it; do not substitute a stale local mirror.
- **Workspace:** Skills remain workspace-scoped. Use each result's workspace; choose a writable one when creating.

If OAuth is unavailable, create a manual SkilzVolt connection key and keep it in the client's private credential or header configuration. Create it on your SkilzVolt Agents page. Treat it like a password: copy it once, store it only in the client's private credential or header setting, and never paste it into chat, an instruction file, or a server URL.

• Server settings: use https://app.skilzvolt.com/mcp, choose Streamable HTTP, and set Authorization: Bearer <SkilzVolt connection key> as the header.
• Codex CLI: codex mcp add skilzvolt --url https://app.skilzvolt.com/mcp --header "Authorization: Bearer <SkilzVolt connection key>"

After the server is connected, use the instruction message in the section above: copy it, paste it into your agent's chat box, and send it. If you lose the key, revoke that connection and create a new one.

Fast, metadata-only bootstrap

skills_catalogue returns authorised names and descriptions plus a catalogue_revision, which the bridge can pass later as known_revision; it does not return full skill content. An OpenClaw bridge can privately retain skill ID, workspace, current version ID, and update time while showing the model a compact - name: description [SkilzVolt] line. Full content is read only through skills_get.

The snapshot is short-lived and bound to the connection. Its revision covers live authorisation and metadata, same-name skills remain distinct by workspace, and it is not a synchronised local mirror. A failed refresh must be reported as degraded mode.

Proposal review lifecycle

Writes create governed proposals and never overwrite approved content. Proposal review data is prepared asynchronously: an edit may briefly report a processing word diff, while a new skill has no earlier version to compare.

1. skills_proposal_status — Check the durable proposal status and word_diff_status.
2. skills_get_proposal_diff — Read the review package and word diff when its status is ready.

Use word_diff_status to distinguish a diff that is processing, ready, not applicable to a new skill, or unavailable. A word_diff: null response does not mean an edit is empty.