Skip to content

Add agent-iteration guidance and the ElevenLabs hosted MCP to the agents skill - #3

Draft
AngeloGiacco wants to merge 1 commit into
mainfrom
claude/architect-plugin-setup-20qrg0
Draft

Add agent-iteration guidance and the ElevenLabs hosted MCP to the agents skill#3
AngeloGiacco wants to merge 1 commit into
mainfrom
claude/architect-plugin-setup-20qrg0

Conversation

@AngeloGiacco

Copy link
Copy Markdown
Collaborator

What

Adds agent-iteration guidance to the agents skill and wires up the ElevenLabs hosted MCP.

  • mcp.json: adds the elevenlabs-hosted server (https://api.elevenlabs.io/v1/mcp), which exposes agent-building operations as direct tools authenticated per user over OAuth, alongside the existing local elevenlabs-mcp server.
  • skills/agents/references/iterating.md (new): the playbook for changing an agent that already exists, explore before change, safe partial update_agent patches, editing large string fields, unified tool CRUD, branches/versions/launches, procedures draft-then-publish, the three test types, and durable facts via prompt/KB. Written against the hosted MCP tool names, with the SDK/xi-api-key equivalent noted.
  • skills/agents/SKILL.md: adds an "Iterating on an Existing Agent" section and a reference link.
  • README.md: documents both MCP servers.

Why

Building an agent from scratch is already covered by the agents skill; iterating on an existing one (branches, safe edits, tools, tests, procedures) has its own reliability rules. This makes that guidance available to any coding agent using the plugin, in Claude Code, Cursor, and Codex.

The tools referenced here are served by the ElevenLabs hosted MCP (backed by the corresponding server-side change in elevenlabs/xi).

Written by claude-opus-4-8, using Claude Code.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant