@x12i/memorix-skills (skill runner)

This package only runs a skill:

1. Resolve skill → {agentDir}/processing/<skill>.md, then auditing/, then code/
2. Render with Rendrix (records → {{input}}, optional catalogs → catalogs)
3. Call dispatcher.run({ ...request, prompt })
4. Return the AiDispatcherResponse unchanged

It does not orchestrate pipelines, land records, or call Nexum.

Internal actions

agents/system/internal/{key}.md is a helper call, not a sequence step. This package ships that folder. A system folder placed beside a running pack replaces the shipped file with the same key. A real pack's internal/ folder is ignored.

Package agents and blocks

agents/ in this package is the published agent root: foundation, system, and the samples record-brief, evidence-summary, and tryout. Real packs stay outside this package. A skill key is loaded from the running pack, then foundation, then system. {{block:name.md}} inserts blocks/name.md before Rendrix fills the other tokens. The same order applies. See agents/BLOCKS.md.

const response = await skills.runInternal({
  action: "audit-prompt",
  input: { prompt: "Write the note." },
});

listInternalActions(skillsDir) and renderInternalAction(action, input) prepare the call without sending it.

Usage

import { createMemorixSkills } from "@x12i/memorix-skills";

const skills = createMemorixSkills({
  skillsDir: "/path/to/.memorix-agents/neo-networks", // one agent directory; default uses AGENTS_DIR or SKILLS_DIR
  dispatcher: {
    openrouter: { apiKey: process.env.OPENROUTER_API_KEY },
  },
});

const response = await skills.run({
  skill: "echo-records",
  records: [{ recordId: "map-1", data: { name: "Example" } }],
  request: {
    provider: "openrouter",
    model: "openai/gpt-5.2",
  },
});

console.log(response.text);

Optional Memorix HTTP client: skills.memorix (station env / options). Unused by run.