feat: require file locations in all EXPLORATORY tier agent outputs

Enhanced all agentic tool EXPLORATORY prompts (Massive tier) to mandate
file location citations for every code element mentioned in analysis.

Changes to 6 prompt files:
- code_search_prompts.rs: Added FILE LOCATIONS REQUIRED as rule #3
- dependency_analysis_prompts.rs: Added to CRITICAL RULES section
- call_chain_prompts.rs: Added MANDATORY FILE LOCATION REQUIREMENT
- architecture_analysis_prompts.rs: Added after YOUR TASK section
- context_builder_prompts.rs: Added after YOUR MISSION section
- semantic_question_prompts.rs: Added after CRITICAL RULE
- agentic_api_surface_prompts.rs: Added after ZERO HEURISTICS RULE

Required format: `ComponentName in path/to/file.rs:line_number`
Example: "ConfigLoader in src/config/loader.rs:42" NOT "ConfigLoader"

This ensures agents receive file paths they can use to drill down into
specific code when needed, enabling better follow-up analysis and code
navigation.

Tool results already contain location data (file_path, start_line) - now
agents are explicitly required to extract and include it in all outputs.

Author: jruokola

crates/codegraph-mcp/src/agentic_api_surface_prompts.rs

@@ -234,6 +234,9 @@ YOUR TASK: Conduct exhaustive API ecosystem analysis including maximum-depth dep
 
 ZERO HEURISTICS RULE: Make NO assumptions about what makes a "good" or "bad" API. Only report factual, measurable graph data from tool outputs.
 
+MANDATORY FILE LOCATION REQUIREMENT:
+For EVERY API function/method/endpoint mentioned, ALWAYS include file location from tool results in format: `APIName in path/to/file.rs:line`. Example: "POST /api/users in src/api/users.rs:23" NOT just "POST /api/users".
+
 AVAILABLE TOOLS:
 1. get_transitive_dependencies(node_id, edge_type, depth) - Follow dependency edges recursively (max depth 10)
 2. detect_circular_dependencies(edge_type) - Find bidirectional dependency cycles across entire codebase

crates/codegraph-mcp/src/architecture_analysis_prompts.rs

@@ -138,6 +138,9 @@ pub const ARCHITECTURE_ANALYSIS_EXPLORATORY: &str = r#"You are a principal archi
 
 YOUR TASK: Perform complete, multi-dimensional architecture analysis using ONLY structured graph metrics - ZERO heuristics.
 
+MANDATORY FILE LOCATION REQUIREMENT:
+For EVERY component/module/class mentioned in your analysis, ALWAYS include file location from tool results in format: `ComponentName in path/to/file.rs:line`. Example: "ConfigLoader in src/config/loader.rs:42" NOT just "ConfigLoader". Tool results contain location data - extract and use it.
+
 AVAILABLE TOOLS (use extensively):
 1. calculate_coupling_metrics(node_id) - Complete coupling analysis for all significant nodes
 2. detect_circular_dependencies(edge_type) - ALL edge types (Imports, Calls, Uses, Extends, Implements)

crates/codegraph-mcp/src/call_chain_prompts.rs

@@ -253,6 +253,9 @@ pub const CALL_CHAIN_EXPLORATORY: &str = r#"You are a principal execution flow a
 
 OBJECTIVE: Perform exhaustive call chain analysis to completely map execution paths, trace all invocation sequences, analyze control flow complexity, identify architectural patterns, and provide deep insights into code execution behavior across the entire codebase.
 
+MANDATORY FILE LOCATION REQUIREMENT:
+For EVERY function/method/component mentioned in your analysis, ALWAYS include its file location from tool results in format: `FunctionName in path/to/file.rs:line_number`. Example: "process_request in src/handlers/request.rs:145" NOT just "process_request". Tool results contain location data (file_path, start_line) - extract and use it so agents can drill down into specific files.
+
 AVAILABLE TOOLS (use comprehensively):
 
 1. trace_call_chain(from_node, max_depth) - PRIMARY TRACING TOOL

crates/codegraph-mcp/src/code_search_prompts.rs

@@ -300,7 +300,14 @@ CRITICAL RULES (ABSOLUTELY MANDATORY):
    - Format: \"From [tool_name] result, node '[exact_node_id]' showed [specific_metric]\"
    - Cross-reference node IDs across multiple tool calls for validation
 
-3. STRUCTURED REASONING CHAIN:
+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. STRUCTURED REASONING CHAIN:
    - Build comprehensive reasoning chains across all tool calls
    - Explicitly connect each tool call to findings from previous calls
    - Cross-validate findings by approaching from multiple angles

crates/codegraph-mcp/src/context_builder_prompts.rs

@@ -174,6 +174,9 @@ pub const CONTEXT_BUILDER_EXPLORATORY: &str = r#"You are a code context builder
 YOUR MISSION:
 Build EXHAUSTIVE, ARCHITECTURALLY COMPLETE context for code understanding or generation. Leave no stone unturned - explore every facet of the codebase relevant to the query.
 
+MANDATORY FILE LOCATION REQUIREMENT:
+For EVERY code element mentioned, ALWAYS include file location from tool results in format: `Name in path/to/file.rs:line`. Example: "parse_config in src/config/parser.rs:89" NOT just "parse_config".
+
 AVAILABLE TOOLS:
 1. get_transitive_dependencies(node_id, edge_type, depth) - Map what this code needs
 2. detect_circular_dependencies(edge_type) - Find dependency cycles

crates/codegraph-mcp/src/dependency_analysis_prompts.rs

@@ -399,6 +399,7 @@ PHASE 7: COMPREHENSIVE SYNTHESIS & RECOMMENDATIONS
 CRITICAL RULES (Strict Adherence):
 - ABSOLUTE ZERO HEURISTICS: Every single claim must be backed by tool data
 - EXTRACT ALL NODE IDS: Use exact IDs from ALL tool results
+- FILE LOCATIONS REQUIRED: For EVERY node/component/function mentioned in your analysis, ALWAYS include its file location using data from tool results. Format: `ComponentName in src/path/file.rs:line_number`. Example: "WorkflowEngine in src/workflow/engine.rs:42" not just "WorkflowEngine"
 - MULTI-DIMENSIONAL COMPLETENESS: Cover all edge types, all depths, all hubs
 - STATISTICAL RIGOR: Calculate distributions, correlations, clustering
 - EXHAUSTIVE ENUMERATION: List ALL circular dependencies, ALL hubs, ALL outliers

crates/codegraph-mcp/src/semantic_question_prompts.rs

@@ -323,6 +323,9 @@ pub const SEMANTIC_QUESTION_EXPLORATORY: &str = r#"You are a principal code anal
 CRITICAL RULE - ZERO HEURISTICS:
 You MUST answer questions using EXCLUSIVELY the graph tools and their empirical results.
 
+MANDATORY FILE LOCATION REQUIREMENT:
+For EVERY code element mentioned in your answer, ALWAYS include file location from tool results in format: `Name in path/to/file.rs:line`. Example: "authenticate in src/auth/handler.rs:156" NOT just "authenticate".
+
 FORBIDDEN REASONING:
 1. General programming knowledge or idioms
 2. Naming conventions, prefixes, or suffixes