Installation

To install this module, you need to:

Download the module and add it to your Odoo addons folder. Afterward, log on to your Odoo server and go to the Apps menu. Trigger the debug mode and update the list by clicking on the "Update Apps List" link. Now install the module by clicking on the install button. Requires muk_ai and muk_mcp.

Upgrade

To upgrade this module, you need to:

Download the module and add it to your Odoo addons folder. Restart the server and log on to your Odoo server. Select the Apps menu and upgrade the module by clicking on the upgrade button.

What's in the box

• muk_ai.skill model — name, label, description, markdown body, attached resources (ir.attachment), agent scoping (specific agents or global), prompt revision history via the reusable muk_ai.revision.mixin from muk_ai.
• invoke_skill MCP tool — agent-callable, returns the skill body plus a resource manifest with uri entries the agent can fetch via muk_mcp's generic read_resource.
• System-prompt addendum — every visible skill is listed by technical name + one-line description inside an <available_skills> block appended to the agent's system prompt.
• Slash command in chat — typing /<skill_name> in the composer fires the skill directly, mirroring the same invoke_skill flow the LLM uses but without an LLM round-trip.
• Manage menu link — MuK AI > Skills (and /odoo/ai-skills).

Building a skill

Open MuK AI > Skills and create a record:

• Technical Name — lowercase identifier matching [a-z][a-z0-9_]*, used for the /name slash command and in the LLM-facing addendum (e.g. quote_followup, lead_brief, summarize_thread).
• Label — human-readable name shown in lists, the chat menu, and the manage view. Optional; falls back to a title-cased version of the technical name.
• Description — one-line description shown to the LLM in the system-prompt addendum. The LLM uses this to decide when to invoke the skill autonomously.
• Body — markdown instructions returned when the skill is invoked. Treated as additional system-prompt content the agent should follow for the duration of the task. Edited with the same code widget as agent system prompts.
• Resources — attached files. Each one becomes a manifest entry with a stable odoo://attachment/<id> URI; the agent fetches the bytes by passing that URI to read_resource.
• Agents — leave empty to make the skill global, or pick one or more agents to scope visibility. Sessions only see skills whose agent scope matches (or skills with no scope).
• Active / Sequence — standard Odoo flags; sequence drives the order in the addendum and the slash-command popover.

Body edits snapshot the prior value into the skill's prompt_history JSON column. The History stat button on the skill form opens a side-by-side dialog listing every prior revision with author and timestamp; restoring rolls back to that revision.

Agent invocation

When a session starts, muk_ai_skills walks the visible-skill set for the session's agent and appends an <available_skills> block to the system prompt. If the agent calls invoke_skill(skill_name='quote_followup'), the tool returns:

{
  "name": "quote_followup",
  "label": "Quote Follow-up",
  "body": "## Steps\n1. Look up the sale.order by ref or domain…",
  "resources": [
    {
      "name": "tone_guide.md",
      "uri": "odoo://attachment/42",
      "mimetype": "text/markdown"
    }
  ]
}

The agent treats the body as fresh instructions and reads any of the listed resources via read_resource(uri='odoo://attachment/42') — no new tool needed for that, the URI handler is in muk_mcp.

User invocation

Type / in the composer to open the slash-command popover. Built-in commands (/help, /clear, /compact, /unpin) are merged with the visible skills for the session. Picking /quote_followup (or typing it and hitting Enter) calls invoke_skill_from_chat on the session, which:

1. Records a tool_call and matching tool_result in the session event log (so the chat shows the skill invocation as a normal tool card with the skill body and manifest as the result).
2. Extends the underlying conversation with a synthetic function_call + function_call_output pair, so on the next model turn the agent sees the skill body as if it had called invoke_skill. The model then proceeds with whatever the skill instructs.

Skills outside the session's agent scope are not listed and not invocable, even by exact name.

Scoping

• Leave Agents empty on a skill to make it visible to every session (global skill).
• Pick one or more agents to restrict the skill to those agents only. A session with no agent_id only sees global skills.
• The visibility filter is applied in both directions: the addendum the LLM sees, and the invoke_skill permission check.

Credits