feat: add FILE LOCATIONS REQUIRED rule to all EXPLORATORY tier prompts

Adds a new critical rule (rule 3) to all EXPLORATORY (Massive tier) prompts
requiring that file locations be included for every node/function/class/component
mentioned in analysis results. This enables agents to drill down into specific
files when needed during code analysis.

Updated prompts:
- call_chain_prompts.rs (CALL_CHAIN_EXPLORATORY)
- architecture_analysis_prompts.rs (ARCHITECTURE_ANALYSIS_EXPLORATORY)
- context_builder_prompts.rs (CONTEXT_BUILDER_EXPLORATORY)
- semantic_question_prompts.rs (SEMANTIC_QUESTION_EXPLORATORY)
- agentic_api_surface_prompts.rs (API_SURFACE_EXPLORATORY)

All existing rules have been renumbered to accommodate the new rule #3.

Author: jruokola

crates/codegraph-mcp/src/agentic_api_surface_prompts.rs

@@ -347,12 +347,18 @@ When analysis is complete:
 }
 
 CRITICAL RULES:
-- Extract ALL node IDs from previous tool results - never invent them
-- Report exact metric values and counts without any qualitative language
-- Use maximum depth parameters to achieve complete ecosystem mapping
-- Cross-reference ALL analyses for comprehensive characterization
-- Build complete dependency/impact graphs at multiple depths
-- Calculate ecosystem-wide statistics where applicable
-- Stay within 20 tool calls maximum (plan strategically)
-- Provide final analysis only when you have exhaustive API ecosystem coverage
-- Focus on measurable quantities: counts, depths, degrees, coupling metrics, impact radii"#;
+1. Extract ALL node IDs from previous tool results - never invent them
+3. FILE LOCATIONS REQUIRED:
+   - For EVERY node/function/class/component mentioned, ALWAYS include its file location from tool results
+   - Format: `ComponentName in path/to/file.rs:line_number` or `ComponentName (path/to/file.rs:line_number)`
+   - Example: "ConfigLoader in src/config/loader.rs:42" NOT just "ConfigLoader"
+   - Tool results contain location data (file_path, start_line) - extract and use it
+   - This allows agents to drill down into specific files when needed
+4. Report exact metric values and counts without any qualitative language
+5. Use maximum depth parameters to achieve complete ecosystem mapping
+6. Cross-reference ALL analyses for comprehensive characterization
+7. Build complete dependency/impact graphs at multiple depths
+8. Calculate ecosystem-wide statistics where applicable
+9. Stay within 20 tool calls maximum (plan strategically)
+10. Provide final analysis only when you have exhaustive API ecosystem coverage
+11. Focus on measurable quantities: counts, depths, degrees, coupling metrics, impact radii"#;

crates/codegraph-mcp/src/architecture_analysis_prompts.rs

@@ -165,7 +165,19 @@ When complete - provide exhaustive report with:
 - Cross-dimensional patterns and statistical health metrics
 - Exhaustive objective synthesis
 
-CONSTRAINTS:
+CRITICAL RULES:
+1. ZERO HEURISTICS: Every claim must be based on tool output data
+2. EXHAUSTIVE NODE ID TRACKING: Extract and reference all node IDs from tool results
+3. FILE LOCATIONS REQUIRED:
+   - For EVERY node/function/class/component mentioned, ALWAYS include its file location from tool results
+   - Format: `ComponentName in path/to/file.rs:line_number` or `ComponentName (path/to/file.rs:line_number)`
+   - Example: "ConfigLoader in src/config/loader.rs:42" NOT just "ConfigLoader"
+   - Tool results contain location data (file_path, start_line) - extract and use it
+   - This allows agents to drill down into specific files when needed
+4. NEVER FABRICATE: Do not invent component names or relationships
+5. MULTI-DIMENSIONAL ANALYSIS: Use all available metrics systematically
+
+OPERATIONAL CONSTRAINTS:
 - MAX 20 STEPS - exhaustive coverage
 - Statistical rigor across all dimensions
 - Multi-edge type exhaustive analysis

crates/codegraph-mcp/src/call_chain_prompts.rs

@@ -360,13 +360,19 @@ When exhaustive analysis is complete:
 }
 
 CRITICAL REQUIREMENTS:
-- ZERO HEURISTICS: Every single claim must cite specific tool output
-- EXHAUSTIVE NODE ID TRACKING: Extract and reference all node IDs from results
-- NEVER FABRICATE: Do not invent function names, IDs, or relationships
-- INCREMENTAL BUILDING: Each step builds on concrete previous results
-- QUANTITATIVE RIGOR: Include metrics, counts, statistics from tool data
-- COMPLETE CITATIONS: Reference specific tool calls and output fields
-- COMPREHENSIVE COVERAGE: Trace ALL significant execution paths, not just main path
+1. ZERO HEURISTICS: Every single claim must cite specific tool output
+2. EXHAUSTIVE NODE ID TRACKING: Extract and reference all node IDs from results
+3. FILE LOCATIONS REQUIRED:
+   - For EVERY node/function/class/component mentioned, ALWAYS include its file location from tool results
+   - Format: `ComponentName in path/to/file.rs:line_number` or `ComponentName (path/to/file.rs:line_number)`
+   - Example: "ConfigLoader in src/config/loader.rs:42" NOT just "ConfigLoader"
+   - Tool results contain location data (file_path, start_line) - extract and use it
+   - This allows agents to drill down into specific files when needed
+4. NEVER FABRICATE: Do not invent function names, IDs, or relationships
+5. INCREMENTAL BUILDING: Each step builds on concrete previous results
+6. QUANTITATIVE RIGOR: Include metrics, counts, statistics from tool data
+7. COMPLETE CITATIONS: Reference specific tool calls and output fields
+8. COMPREHENSIVE COVERAGE: Trace ALL significant execution paths, not just main path
 
 EXECUTION FLOW ANALYSIS FOCUS (Comprehensive):
 ✓ Complete call chain topology from all entry points

crates/codegraph-mcp/src/context_builder_prompts.rs

@@ -193,12 +193,19 @@ CONTEXT BUILDING STRATEGY (Exploratory):
 - Iterative refinement: explore → analyze → explore deeper based on findings
 
 CRITICAL CONSTRAINTS:
-- ZERO HEURISTICS: Use only structured data from graph tools
-- Make 15-20+ tool calls for exhaustive coverage
-- Multi-pass strategy: broad discovery → deep exploration → synthesis
-- Explore ALL edge types systematically
-- Build complete architectural map
-- Connect findings across different analysis dimensions
+1. ZERO HEURISTICS: Use only structured data from graph tools
+2. EXHAUSTIVE NODE ID TRACKING: Extract and reference all node IDs from tool results
+3. FILE LOCATIONS REQUIRED:
+   - For EVERY node/function/class/component mentioned, ALWAYS include its file location from tool results
+   - Format: `ComponentName in path/to/file.rs:line_number` or `ComponentName (path/to/file.rs:line_number)`
+   - Example: "ConfigLoader in src/config/loader.rs:42" NOT just "ConfigLoader"
+   - Tool results contain location data (file_path, start_line) - extract and use it
+   - This allows agents to drill down into specific files when needed
+4. Make 15-20+ tool calls for exhaustive coverage
+5. Multi-pass strategy: broad discovery → deep exploration → synthesis
+6. Explore ALL edge types systematically
+7. Build complete architectural map
+8. Connect findings across different analysis dimensions
 
 RESPONSE FORMAT (REQUIRED JSON):
 {

crates/codegraph-mcp/src/semantic_question_prompts.rs

@@ -324,18 +324,25 @@ CRITICAL RULE - ZERO HEURISTICS:
 You MUST answer questions using EXCLUSIVELY the graph tools and their empirical results.
 
 FORBIDDEN REASONING:
-- General programming knowledge or idioms
-- Naming conventions, prefixes, or suffixes
-- Common patterns or best practices not evidenced in graph
-- Domain assumptions or typical behaviors
-- Any claim not directly supported by graph relationships
+1. General programming knowledge or idioms
+2. Naming conventions, prefixes, or suffixes
+3. Common patterns or best practices not evidenced in graph
+4. Domain assumptions or typical behaviors
+5. Any claim not directly supported by graph relationships
 
 REQUIRED REASONING:
-- Concrete node IDs and edge relationships discovered through tools
-- Quantitative metrics from graph analysis
-- Statistical patterns across multiple tool calls
-- Structural properties of the dependency/call graph
-- Empirical evidence from comprehensive tool exploration
+1. Concrete node IDs and edge relationships discovered through tools
+2. EXHAUSTIVE NODE ID TRACKING: Extract and reference all node IDs from tool results
+3. FILE LOCATIONS REQUIRED:
+   - For EVERY node/function/class/component mentioned, ALWAYS include its file location from tool results
+   - Format: `ComponentName in path/to/file.rs:line_number` or `ComponentName (path/to/file.rs:line_number)`
+   - Example: "ConfigLoader in src/config/loader.rs:42" NOT just "ConfigLoader"
+   - Tool results contain location data (file_path, start_line) - extract and use it
+   - This allows agents to drill down into specific files when needed
+4. Quantitative metrics from graph analysis
+5. Statistical patterns across multiple tool calls
+6. Structural properties of the dependency/call graph
+7. Empirical evidence from comprehensive tool exploration
 
 If graph structure is insufficient to answer definitively, provide:
 1. What IS known from graph evidence