fix: address review findings for accuracy

- Fix plugin discovery docs: uses __all__ exports, not @tool scanning
- Fix widget markdown claims: no syntax highlighting or strikethrough
- Add source_repo field to DocSource schema reference table
- Fix widget API response: tool_calls not session_id
- Add note about cf_turnstile_response handled by Worker proxy
- Add missing widget config options: repoUrl, repoName, pageContext*
- Fix communityId validation description to match actual regex
- Fix extensions loading order to match actual init sequence

Author: neuromechanist

docs/osa/deployment/widget.md

@@ -30,8 +30,12 @@ The widget appears as a chat bubble in the bottom-right corner of the page.
 | `storageKey` | string | Auto-derived | localStorage key for chat history |
 | `turnstileSiteKey` | string | `null` | Cloudflare Turnstile site key |
 | `showExperimentalBadge` | boolean | `true` | Show beta/experimental badge |
+| `repoUrl` | string | OSA GitHub URL | URL for the "Powered by" footer link |
+| `repoName` | string | `'Open Science Assistant'` | Display name in footer |
 | `allowPageContext` | boolean | `true` | Show page context toggle |
 | `pageContextDefaultEnabled` | boolean | `true` | Default state of page context |
+| `pageContextStorageKey` | string | `'osa-page-context-enabled'` | localStorage key for page context preference |
+| `pageContextLabel` | string | `'Share page URL...'` | Label text for the page context checkbox |
 | `fullscreen` | boolean | `false` | Open chat in fullscreen mode |
 
 ### Minimal Configuration
@@ -99,10 +103,10 @@ Users can open the chat in a separate browser window for a larger workspace. The
 Assistant responses support full Markdown rendering including:
 
 - Tables
-- Code blocks with syntax highlighting
+- Code blocks with copy button
 - Lists (ordered and unordered)
 - Links
-- Bold, italic, strikethrough
+- Bold and italic
 
 ## Environment Detection
 
@@ -153,17 +157,18 @@ The widget communicates with two backend endpoints:
   "page_context": {
     "url": "https://hedtags.org/docs/getting-started",
     "title": "Getting Started - HED"
-  },
-  "cf_turnstile_response": "token..."
+  }
 }
 ```
 
+The `cf_turnstile_response` field is also sent by the widget when Turnstile is configured, but this is consumed by the Cloudflare Worker proxy, not the backend API.
+
 ### Response Format
 
 ```json
 {
   "answer": "HED (Hierarchical Event Descriptors) is...",
-  "session_id": "abc123"
+  "tool_calls": []
 }
 ```
 
@@ -203,7 +208,7 @@ Each community page auto-configures the widget with the appropriate settings.
 
 - Check browser console for JavaScript errors
 - Verify the script URL is accessible
-- Ensure `communityId` contains only lowercase letters, numbers, and hyphens
+- Ensure `communityId` contains only letters, numbers, hyphens, and underscores (the backend registry uses kebab-case IDs)
 
 **"Offline" status:**
 

docs/osa/registry/extensions.md

@@ -93,13 +93,20 @@ extensions:
         - search_examples
 ```
 
-If `tools` is omitted, all `@tool`-decorated functions in the module are loaded:
+If `tools` is omitted, all names exported in the module's `__all__` list that are valid tool objects are loaded:
 
 ```yaml
 extensions:
   python_plugins:
     - module: src.assistants.my-tool.tools
-      # All @tool functions loaded
+      # All tools listed in __all__ are loaded
+```
+
+Your module must define `__all__` listing the tools to export:
+
+```python
+# src/assistants/my-tool/tools.py
+__all__ = ["validate_config", "search_examples"]
 ```
 
 ### Example: HED Tools
@@ -174,13 +181,14 @@ Exactly one of `command` or `url` must be provided for each server.
 
 When a `CommunityAssistant` is created, tools are loaded in this order:
 
-1. **Auto-generated tools** from YAML config:
-    - `retrieve_{community}_docs` - Document retrieval
+1. **Knowledge tools** from YAML config:
     - `list_{community}_recent` - Recent GitHub activity
     - `search_{community}_discussions` - GitHub search
     - `search_{community}_papers` - Academic paper search
-    - `fetch_current_page` - Page context (if enabled)
-2. **Python plugin tools** from `extensions.python_plugins`
-3. **MCP server tools** from `extensions.mcp_servers` (when implemented)
+2. **Documentation retrieval** - `retrieve_{community}_docs`
+3. **Page context** - `fetch_current_page` (if `enable_page_context: true`)
+4. **Additional tools** passed programmatically (if any)
+5. **Python plugin tools** from `extensions.python_plugins`
+6. **MCP server tools** from `extensions.mcp_servers` (when implemented)
 
 All tools are available to the LLM simultaneously. The system prompt should guide the LLM on when to use each tool.

docs/osa/registry/schema-reference.md

@@ -77,6 +77,7 @@ List of documentation sources. Each entry creates a retrievable document.
 | `preload` | boolean | No | `false` | Embed in system prompt at startup |
 | `category` | string | No | `general` | Category for organization |
 | `type` | enum | No | `html` | Format: `sphinx`, `mkdocs`, `html`, `markdown`, `json` |
+| `source_repo` | string | No | `null` | GitHub repo for raw markdown sources (e.g., 'org/repo') |
 | `description` | string | No | `null` | Short description of document content |
 
 ### Preloaded vs On-Demand