Tighten system prompt and tool definitions

Prompt changes (native.md):
- Remove duplicate 'Temporary files' section
- Consolidate 'remember' instructions into single authoritative location
- Remove motivational 'Benefits' list from Plan Mode
- Add 'Code Search Tool Selection' guidance (code_search vs rg)

Tool changes (tool_definitions.rs, tool_dispatch.rs):
- Remove screenshot tool (webdriver_screenshot remains)
- Remove coverage tool
- Reduce plan_write description from 22 lines to 1 line
- Update tool count tests (16 -> 14 core tools)

Net result: ~6 lines removed from prompt, ~56 lines removed from
tool definitions, clearer tool selection guidance added.

crates/g3-core/src/tool_definitions.rs

@@ -166,42 +166,6 @@ fn create_core_tools(exclude_research: bool) -> Vec<Tool> {
                 "required": ["file_path", "diff"]
             }),
         },
-        Tool {
-            name: "screenshot".to_string(),
-            description: "Capture a screenshot of a specific application window. You MUST specify the window_id parameter with the application name (e.g., 'Safari', 'Terminal', 'Google Chrome'). The tool will automatically use the native screencapture command with the application's window ID for a clean capture. Use list_windows first to identify available windows.".to_string(),
-            input_schema: json!({
-                "type": "object",
-                "properties": {
-                    "path": {
-                        "type": "string",
-                        "description": "Filename for the screenshot (e.g., 'safari.png'). If a relative path is provided, the screenshot will be saved to ~/tmp or $TMPDIR. Use an absolute path to save elsewhere."
-                    },
-                    "window_id": {
-                        "type": "string",
-                        "description": "REQUIRED: Application name to capture (e.g., 'Safari', 'Terminal', 'Google Chrome'). The tool will capture the frontmost window of that application using its native window ID."
-                    },
-                    "region": {
-                        "type": "object",
-                        "properties": {
-                            "x": {"type": "integer"},
-                            "y": {"type": "integer"},
-                            "width": {"type": "integer"},
-                            "height": {"type": "integer"}
-                        }
-                    }
-                },
-                "required": ["path", "window_id"]
-            }),
-        },
-        Tool {
-            name: "coverage".to_string(),
-            description: "Generate a code coverage report for the entire workspace using cargo llvm-cov. This runs all tests with coverage instrumentation and returns a summary of coverage statistics. Requires llvm-tools-preview and cargo-llvm-cov to be installed (they will be auto-installed if missing).".to_string(),
-            input_schema: json!({
-                "type": "object",
-                "properties": {},
-                "required": []
-            }),
-        },
         Tool {
             name: "code_search".to_string(),
             description: "Syntax-aware code search that understands code structure, not just text. Finds actual functions, classes, methods, and other code constructs - ignores matches in comments and strings. Much more accurate than grep for code searches. Supports batch searches (up to 20 parallel) with structured results and context lines. Languages: Rust, Python, JavaScript, TypeScript, Go, Java, C, C++, Racket. Uses tree-sitter query syntax.".to_string(),
@@ -278,27 +242,7 @@ fn create_core_tools(exclude_research: bool) -> Vec<Tool> {
 
     tools.push(Tool {
         name: "plan_write".to_string(),
-        description: r#"Create or update the Plan for this session. The plan must be provided as YAML with the following structure:
+        description: "Create or update the Plan for this session. Provide plan as YAML with plan_id and items array. See system prompt for full schema (items need: id, description, state, touches, checks with happy/negative/boundary). Evidence and notes required when marking done.".to_string(),
-
-- plan_id: Unique identifier for the plan
-- revision: Will be auto-incremented
-- items: Array of plan items, each with:
-  - id: Stable identifier (e.g., "I1")
-  - description: What will be done
-  - state: todo | doing | done | blocked
-  - touches: Array of paths/modules affected
-  - checks:
-      happy: {desc, target} - Normal successful operation
-      negative: [{desc, target}, ...] - Error handling, invalid input (>=1 required)
-      boundary: [{desc, target}, ...] - Edge cases, limits (>=1 required)
-  - evidence: Array of file:line refs, test names (required when done)
-  - notes: Implementation explanation (required when done)
-
-Rules:
-- Keep items ≤ 7 by default
-- All checks required: 1 happy, 1+ negative, 1+ boundary
-- Cannot remove items from an approved plan (mark as blocked instead)
-- Evidence and notes required when marking item as done"#.to_string(),
         input_schema: json!({
             "type": "object",
             "properties": {
@@ -557,10 +501,10 @@ mod tests {
     fn test_core_tools_count() {
         let tools = create_core_tools(false);
         // Core tools: shell, background_process, read_file, read_image,
-        // write_file, str_replace, screenshot, coverage, code_search,
+        // write_file, str_replace, code_search,
         // research, research_status, remember, plan_read, plan_write, plan_approve
-        // (16 total - memory is auto-loaded, only remember tool needed)
+        // (14 total - memory is auto-loaded, only remember tool needed)
-        assert_eq!(tools.len(), 16);
+        assert_eq!(tools.len(), 14);
     }
 
     #[test]
@@ -574,15 +518,15 @@ mod tests {
     fn test_create_tool_definitions_core_only() {
         let config = ToolConfig::default();
         let tools = create_tool_definitions(config);
-        assert_eq!(tools.len(), 16);
+        assert_eq!(tools.len(), 14);
     }
 
     #[test]
     fn test_create_tool_definitions_all_enabled() {
         let config = ToolConfig::new(true, true);
         let tools = create_tool_definitions(config);
-        // 16 core + 15 webdriver = 31
+        // 14 core + 15 webdriver = 29
-        assert_eq!(tools.len(), 31);
+        assert_eq!(tools.len(), 29);
     }
 
     #[test]
@@ -600,8 +544,8 @@ mod tests {
         let tools_with_research = create_core_tools(false);
         let tools_without_research = create_core_tools(true);
         
-        assert_eq!(tools_with_research.len(), 16);
+        assert_eq!(tools_with_research.len(), 14);
-        assert_eq!(tools_without_research.len(), 14);  // research + research_status both excluded
+        assert_eq!(tools_without_research.len(), 12);  // research + research_status both excluded
         
         assert!(tools_with_research.iter().any(|t| t.name == "research"));
         assert!(!tools_without_research.iter().any(|t| t.name == "research"));

crates/g3-core/src/tool_dispatch.rs

@@ -38,8 +38,6 @@ pub async fn dispatch_tool<W: UiWriter>(
         "plan_approve" => plan::execute_plan_approve(tool_call, ctx).await,
 
         // Miscellaneous tools
-        "screenshot" => misc::execute_take_screenshot(tool_call, ctx).await,
-        "coverage" => misc::execute_code_coverage(tool_call, ctx).await,
         "code_search" => misc::execute_code_search(tool_call, ctx).await,
 
         // Research tool

prompts/system/native.md

@@ -14,6 +14,13 @@ IMPORTANT: You must call tools to achieve goals. When you receive a request:
 For shell commands: Use the shell tool with the exact command needed. Always use `rg` (ripgrep) instead of `grep` - it's faster, has better defaults, and respects .gitignore. Avoid commands that produce a large amount of output, and consider piping those outputs to files. Example: If asked to list files, immediately call the shell tool with command parameter "ls".
 If you create temporary files for verification, place these in a subdir named 'tmp'. Do NOT pollute the current dir.
 
+# Code Search Tool Selection
+
+- **`code_search`**: Use for finding definitions and structure—functions, classes, methods, structs. Syntax-aware (ignores matches in comments/strings). Best for "where is X defined?" or "find all implementations of Y".
+- **`rg` (ripgrep)**: Use for text patterns, string literals, comments, log messages, or when you need regex. Best for "find all uses of this error message" or "grep for TODO".
+
+When in doubt: `code_search` for definitions, `rg` for text.
+
 # Task Management with Plan Mode
 
 **REQUIRED for all tasks.**
@@ -30,7 +37,6 @@ Plan Mode is a cognitive forcing system that prevents:
 2. **Approval**: Ask user to approve before starting work ("'approve', or edit plan?"). In non-interactive mode (autonomous/one-shot), plans auto-approve on write.
 3. **Execute**: Implement items, updating plan with `plan_write` to mark progress
 4. **Complete**: When all items are done/blocked, verification runs automatically
-5. **Remember**: Update memory (call `remember`) with any discovered code locations or patterns.
 
 ## Plan Schema
 
@@ -152,18 +158,6 @@ facts:
 3. After all work is complete, write `envelope.yaml` with facts about the completed work
 4. **THEN** call `plan_write` to mark the final item done - verification will check both files
 
-## Benefits
-
-✓ Prevents missed steps
-✓ Makes progress visible
-✓ Helps recover from interruptions
-✓ Forces consideration of edge cases
-✓ Provides audit trail with evidence
-
-# Temporary files
-
-If you create temporary files for verification or investigation, place these in a subdir named 'tmp'. Do NOT pollute the current dir.
-
 # Web Research
 
 When you need to look up documentation, search for resources, find data online, or research a topic to complete your task, use the `research` tool. **Research is asynchronous** - it runs in the background while you continue working.
@@ -221,5 +215,5 @@ This applies whenever you use search tools like `code_search`, `rg`, `grep`, `fi
 
 - Use Markdown formatting for all responses except tool calls.
 - Whenever taking actions, use the pronoun 'I'
-- When you discover features, patterns and code locations, call `remember` to save them.
+- Call `remember` at end of turn if you discovered code locations (see Workspace Memory section).
 - When showing example tool call JSON in prose or code blocks, use the fullwidth left curly bracket `｛` (U+FF5B) instead of `{` to prevent parser confusion.