fix: prevent temp file accumulation during long agent runs

Address three issues reported after overnight AutoForge runs:
1. ~193GB of .node files in %TEMP% from V8 compile caching
2. Stale npm artifact folders on drive root when %TEMP% fills up
3. PNG screenshot files left in project root by Playwright

Changes:
- Widen .node cleanup glob from ".78912*.node" to ".[0-9a-f]*.node"
  to match all V8 compile cache hex prefixes
- Add "node-compile-cache" directory to temp cleanup patterns
- Set NODE_COMPILE_CACHE="" in all subprocess environments (client.py,
  parallel_orchestrator.py, process_manager.py) to disable V8 compile
  caching at the source
- Add cleanup_project_screenshots() to remove stale .png files from
  project directories (feature*-*.png, screenshot-*.png, step-*.png)
- Run cleanup_stale_temp() at server startup in lifespan()
- Add _run_inter_session_cleanup() to orchestrator, called after each
  agent completes (both coding and testing paths)
- Update coding and testing prompt templates to instruct agents to use
  inline (base64) screenshots only, never saving files to disk

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

.claude/templates/coding_prompt.template.md

@@ -90,13 +90,13 @@ Use browser automation tools:
 
 - Navigate to the app in a real browser
 - Interact like a human user (click, type, scroll)
-- Take screenshots at each step
+- Take screenshots at each step (use inline screenshots only -- do NOT save screenshot files to disk)
 - Verify both functionality AND visual appearance
 
 **DO:**
 
 - Test through the UI with clicks and keyboard input
-- Take screenshots to verify visual appearance
+- Take screenshots to verify visual appearance (inline only, never save to disk)
 - Check for console errors in browser
 - Verify complete user workflows end-to-end
 
@@ -194,6 +194,8 @@ Before context fills up:
 
 Use Playwright MCP tools (`browser_*`) for UI verification. Key tools: `navigate`, `click`, `type`, `fill_form`, `take_screenshot`, `console_messages`, `network_requests`. All tools have auto-wait built in.
 
+**Screenshot rule:** Always use inline mode (base64). NEVER save screenshots as files to disk.
+
 Test like a human user with mouse and keyboard. Use `browser_console_messages` to detect errors. Don't bypass UI with JavaScript evaluation.
 
 ---

.claude/templates/testing_prompt.template.md

@@ -31,14 +31,14 @@ For the feature returned:
 1. Read and understand the feature's verification steps
 2. Navigate to the relevant part of the application
 3. Execute each verification step using browser automation
-4. Take screenshots to document the verification
+4. Take screenshots to document the verification (inline only -- do NOT save to disk)
 5. Check for console errors
 
 Use browser automation tools:
 
 **Navigation & Screenshots:**
 - browser_navigate - Navigate to a URL
-- browser_take_screenshot - Capture screenshot (use for visual verification)
+- browser_take_screenshot - Capture screenshot (inline mode only -- never save to disk)
 - browser_snapshot - Get accessibility tree snapshot
 
 **Element Interaction:**
@@ -79,7 +79,7 @@ A regression has been introduced. You MUST fix it:
 
 4. **Verify the fix:**
    - Run through all verification steps again
-   - Take screenshots confirming the fix
+   - Take screenshots confirming the fix (inline only, never save to disk)
 
 5. **Mark as passing after fix:**
    ```
@@ -110,7 +110,7 @@ A regression has been introduced. You MUST fix it:
 All interaction tools have **built-in auto-wait** -- no manual timeouts needed.
 
 - `browser_navigate` - Navigate to URL
-- `browser_take_screenshot` - Capture screenshot
+- `browser_take_screenshot` - Capture screenshot (inline only, never save to disk)
 - `browser_snapshot` - Get accessibility tree
 - `browser_click` - Click elements
 - `browser_type` - Type text