# Agent OS Runtime Notes

## Live Model Configuration

Set these environment variables before running the Agent OS goal APIs:

- `AGENT_OS_MODEL_ENDPOINT` (OpenAI-compatible chat completions endpoint)
- `AGENT_OS_MODEL_NAME` (default: `glm-4.7`)
- `AGENT_OS_MODEL_API_KEY`
- `AGENT_OS_MODEL_TIMEOUT_MS` (optional, default `30000`)

`/api/agent/goals` uses model planning first. If model config is missing or call fails, it falls back to heuristic planning and annotates `run.planning.warning`.

## n8n Production Workflow Mapping

`/api/admin/resources/skills` exposes `n8nProductionWorkflows` and corresponding n8n-backed skill manifests:

- Daily Trending discovery
- Topic discovery
- Multi-source signal aggregation
- Project description embedding refresh
- RAG similarity retrieval
- GitHub stars refresh
- Multi-source project ingest

## Runtime Policy Patch Activation

Policy proposals approved through:

- `POST /api/admin/policies/proposals`
- `POST /api/admin/approvals/:id/decision`

now immediately apply runtime patches for `targetType=agent` (policy overrides, command rules, path rules, enabled resource tags). This ensures approval changes affect subsequent runs.

## Live E2E

Playwright suite:

- `e2e/agent-os-live.spec.ts`

Required env:

- `E2E_ADMIN_KEY` (or `ADMIN_CONSOLE_KEY`)
- one of `AGENT_OS_MODEL_API_KEY` / `BIGMODEL_API_KEY` / `ZHIPU_API_KEY`

Run:

```bash
pnpm test:e2e
```