docs: align provider and test harness guidance

README.md

@@ -77,6 +77,16 @@ Deployment profiles and hardening references:
 
 <details>
 
+<summary><strong>Provider URL parity and CI harness resilience tightened for local LLM defaults and Playwright bootstrap stability</strong></summary>
+
+- Fixed the built-in `Ollama (Local)` provider default so OpenClaw's OpenAI-compatible requests now target the correct `/v1` surface by default, and existing loopback-root overrides are normalized onto the same bounded path instead of failing on `/models` or `/chat/completions` at the daemon root.
+- Added a provider URL contract matrix that pins built-in provider defaults, adapter endpoint assembly, and bounded Ollama normalization in one regression lane so future `LM Studio`, `Ollama`, and custom OpenAI-compatible drift is caught before release.
+- Hardened the shared Playwright harness bootstrap so a single transient `openclaw.js` module-fetch failure in CI is retried once instead of failing the whole UI load, while still surfacing real import/runtime errors as hard test failures.
+
+</details>
+
+<details>
+
 <summary><strong>PNG Info sidebar workflow added with ComfyUI metadata extraction, better large-image handling, and lower-noise operator alerts</strong></summary>
 
 - Added a new `PNG Info` sidebar tab with drag-and-drop, file picker, scoped paste, preview rendering, prompt copy actions, structured summary cards, and raw metadata inspection for saved generation images.
@@ -768,6 +778,9 @@ Notes:
 - Optional: for single-user localhost setups, you can store a provider API key from the Settings tab (UI Key Store (Advanced)).
   - This writes to the server-side secret store (`{STATE_DIR}/secrets.json`).
   - Environment variables always take priority over stored keys.
+- Built-in local-provider defaults use loopback-only OpenAI-compatible URLs:
+  - `Ollama (Local)` -> `http://127.0.0.1:11434/v1`
+  - `LM Studio (Local)` -> `http://localhost:1234/v1`
 
 ### 2 Configure webhook auth (required for `/webhook*`)
 

docs/deploy/local-only.md

@@ -35,7 +35,7 @@ No special configuration is required.
 - **Admin Token**: Not required for loopback-only operations (unless `OPENCLAW_ADMIN_TOKEN` is explicitly set).
 - **Webhooks**: Disabled by default.
 - **Local LLM (optional)**:
-  - Ollama: `http://127.0.0.1:11434`
+  - Ollama: `http://127.0.0.1:11434/v1`
   - LM Studio: `http://localhost:1234/v1`
   - Keep SSRF relax flags disabled:
     - `OPENCLAW_ALLOW_ANY_PUBLIC_LLM_HOST=0`
@@ -58,6 +58,6 @@ No special configuration is required.
 1. Open `http://127.0.0.1:8188` in your browser.
 2. Open the OpenClaw tab in the sidebar.
 3. Go to **Settings** -> **Health**.
-4. If using Ollama, verify `http://127.0.0.1:11434/api/tags` responds.
-5. In **Settings -> LLM**, set provider to `Ollama (Local)` and click **Load Models**.
+4. If using Ollama, verify the daemon is reachable first via the native Ollama health/list surface `http://127.0.0.1:11434/api/tags`.
+5. In **Settings -> LLM**, set provider to `Ollama (Local)`, leave **Base URL** empty to use the built-in `http://127.0.0.1:11434/v1` default (or set that exact loopback URL explicitly), and click **Load Models**.
 6. All checks should be green.

docs/frontend_ux_walkthrough.md

@@ -91,5 +91,7 @@ If `assist_streaming` is unavailable or the stream transport degrades, Planner/R
 - Run: `npm test`
 - Tests live in: `tests/e2e/specs/`
 - Harness: `tests/e2e/test-harness.html` (mocks ComfyUI core + basic OpenClaw API calls)
+- Harness bootstrap now retries one transient `openclaw.js` module-fetch failure before surfacing a hard load error, so CI-only first-request flakiness does not get misreported as a permanent sidebar failure.
 - Web helper/self-test harness: `web/tests/e2e-harness.html` (includes frontend helper and wrapper idempotence checks)
 - Desktop host parity lane: `tests/e2e/specs/desktop_host_parity.spec.js` verifies standalone vs desktop host evidence separately and covers both sidebar and Remote Admin host-sensitive behavior under the shared harness shims.
+- When investigating suspected harness flakes locally, prefer `npm run test:stress -- <spec>` so the same shared bootstrap path is exercised repeatedly without changing the default `npm test` contract.

docs/release/config_secrets_contract.md

@@ -62,6 +62,9 @@ Multi-tenant note:
 
 Notes:
 - Local providers (`ollama`, `lmstudio`) are loopback-only by design and should use `localhost` / `127.0.0.1` / `::1`.
+- Built-in local-provider defaults are OpenAI-compatible URLs:
+  - `ollama` -> `http://127.0.0.1:11434/v1`
+  - `lmstudio` -> `http://localhost:1234/v1`
 - Local loopback provider targets do not require enabling insecure SSRF flags.
 - `OPENCLAW_LLM_ALLOWED_HOSTS` does not allow private/reserved IPs; those still require `OPENCLAW_ALLOW_INSECURE_BASE_URL=1`.
 - The same insecure override applies to config-save validation, `/openclaw/llm/models`, and outbound provider requests.

docs/release/operator_ux_contracts.md

@@ -50,7 +50,8 @@ interface BannerStatus {
 - **Monitoring**: Polls `/health` every 10s.
 - **Triggers**: Checks `stats.observability.total_dropped > 0`.
 - **Display**: Simple DOM injection of canonical `.openclaw-banner` markup; legacy `.moltbot-banner` compatibility selectors remain available through centralized runtime aliasing.
-- **Limitations**: No connectivity state handling, no 'info'/'success' states, simplistic dedupe.
+- **Connectivity posture**: Queue-monitor disconnect warnings should tolerate initial sidebar/bootstrap races and only escalate after bounded repeated failure or post-healthy disconnect evidence, so transient startup misses do not become durable incident noise.
+- **Limitations**: No 'info'/'success' states, simplistic dedupe.
 
 ## 1.1 Notification Center (F66)
 

docs/troubleshooting.md

@@ -61,6 +61,10 @@ Checklist:
 3. In OpenClaw Settings:
    - Provider: `Ollama (Local)` or `LM Studio (Local)`
    - Base URL: leave empty to use the provider default, or set a loopback URL explicitly
+   - Provider defaults:
+     - `Ollama (Local)` -> `http://127.0.0.1:11434/v1`
+     - `LM Studio (Local)` -> `http://localhost:1234/v1`
+   - If an older saved Ollama URL is still set to `http://127.0.0.1:11434`, update it to `/v1` or clear the field so the built-in default can be applied
 4. Keep these flags disabled:
    - `OPENCLAW_ALLOW_ANY_PUBLIC_LLM_HOST=0`
    - `OPENCLAW_ALLOW_INSECURE_BASE_URL=0`