Ouroboros Config

/ouroboros:ouroboros-config

Settings for ~/.ouroboros/config.yaml: per-stage runtime/model selects, global runtime + LLM backend, install badges for missing CLIs, and env-override warnings.

Usage

ooo config
/ouroboros:ouroboros-config

Trigger keywords: "ooo config", "open settings", "configure ouroboros", "change model", "change agent"

Instructions

Pick the branch that matches where you (the agent) are running. The decisive question: can the user open a browser pointed at this machine?

Branch A — local harness (Claude Code / Codex on the user's own machine)

1. Launch in the background (the command serves until stopped):

   if command -v ouroboros >/dev/null 2>&1; then
     ouroboros config
   else
     uvx --python '>=3.12' --from 'ouroboros-ai[tui]' ouroboros config
   fi
   

   The command detects the non-interactive context itself and serves the settings app over a local web server, auto-opening the user's browser. The uvx fallback is required for a Marketplace-plugin-only install, where the MCP server exists but ouroboros is not on PATH. In a development checkout use uv run ouroboros config.

2. Relay the http://localhost:<port> line from the output so the user can open it manually if the browser did not pop up.

3. Tell the user: edit → Save → then ask you to stop the server. Remind them a running MCP server may need a reconnect to pick up backend changes. Tell them they can reopen these settings any time with ooo config; saving a model choice never locks it permanently.

Branch B — remote host the user can reach over the network (SSH box, home server)

The user cannot see a browser opened here, but may be able to reach this host. Serve without auto-open and hand over the URL:

ouroboros config --web --host 0.0.0.0 --no-browser

Relay the printed URL with this host's address substituted, plus the SSH tunnel fallback the command prints (ssh -L <port>:localhost:<port> <this-host>).

Branch C — chat gateway, no browser path at all (e.g. hermes driven from Discord)

Do NOT start a server nobody can reach. Drive the same settings conversationally over the scriptable surface:

1. Show the current state:

   ouroboros config show
   

2. Present the user a short menu in chat — default agent, per-stage agents, per-stage models — with the current values, and ask what to change.

3. Apply each choice with the validated setter (same write path as the GUI):

   ouroboros config set orchestrator.runtime_backend <agent>
   ouroboros config set orchestrator.runtime_profile.stages.<interview|execute|evaluate|reflect> <agent>
   ouroboros config set clarification.default_model <model>        # interview & seed
   ouroboros config set execution.default_model <model>            # execute
   ouroboros config set evaluation.semantic_model <model>          # evaluate
   ouroboros config set resilience.reflect_model <model>           # reflect
   ouroboros config set llm.backend <backend>                      # internal LLM calls
   

4. Confirm with ouroboros config show and summarize what changed.

If a set is rejected, relay the validation error verbatim — it lists the valid keys/values.

All branches

If the command fails with a missing-dependency hint, relay it verbatim (pip install 'ouroboros-ai[tui]'). Scriptable edits always remain on ouroboros config show|set|backend|init|validate.

End your final message with the state breadcrumb footer (RFC #1392), e.g.:

◆ Settings GUI serving at <url> → next: Save in browser, then stop the server
◆ Config updated via chat (<keys>) → next: reconnect MCP if the backend changed

RFC #1392 State Breadcrumb Footer

Your final response MUST end with exactly one breadcrumb footer line:

◆ <current state> → next: <recommended action>

Derive <current state> from live session state via ouroboros_session_status when that MCP projection is available; otherwise derive it from this skill's actual outcome. Never use a linear Step N of M footer because Ouroboros is an evolutionary loop. When the next action is genuinely a choice, list 2-3 honest options in the next: clause. The breadcrumb line must be the last line of the response.