fix: direct agent validation screenshots to .autoforge/validation and gitignore them

.claude/templates/coding_prompt.template.md

@@ -90,7 +90,7 @@ 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 and store them in the `.autoforge/screenshots` directory
 - Verify both functionality AND visual appearance
 
 **DO:**

.claude/templates/testing_prompt.template.md

@@ -31,7 +31,7 @@ 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 and store them in the `.autoforge/screenshots` directory
 5. Check for console errors
 
 Use browser automation tools:

CLAUDE.md

@@ -287,6 +287,7 @@ Projects can be stored in any directory (registered in `~/.autoforge/registry.db
 - `.autoforge/features.db` - SQLite database with feature test cases
 - `.autoforge/.agent.lock` - Lock file to prevent multiple agent instances
 - `.autoforge/allowed_commands.yaml` - Project-specific bash command allowlist (optional)
+- `.autoforge/validation` - Agent produced artifacts when validating work
 - `.autoforge/.gitignore` - Ignores runtime files
 - `CLAUDE.md` - Stays at project root (SDK convention)
 - `app_spec.txt` - Root copy for agent template compatibility

autoforge_paths.py

@@ -43,6 +43,7 @@
 .claude_assistant_settings.json
 .claude_settings.expand.*.json
 .progress_cache
+screenshots/
 """