@@ -39,6 +39,7 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
3939#### ** Structured Output Enforcement with JSON Schemas**
4040- ** JSON schema enforcement** for all 7 agentic tools with ** required file paths**
4141- ** Schema-driven outputs** : LLM cannot return response without file locations
42+ - ** AutoAgents integration** : Uses ` StructuredOutputFormat ` passed via ` AgentDeriveT::output_schema() `
4243- ** New module** : ` codegraph-ai/src/agentic_schemas.rs ` with comprehensive schemas:
4344 - ` CodeSearchOutput ` : analysis + components[ ] + patterns[ ]
4445 - ` DependencyAnalysisOutput ` : analysis + components[ ] + dependencies[ ] + circular_dependencies[ ]
@@ -47,18 +48,23 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
4748 - ` APISurfaceOutput ` : analysis + endpoints[ ] + usage_patterns[ ]
4849 - ` ContextBuilderOutput ` : comprehensive context with all analysis dimensions
4950 - ` SemanticQuestionOutput ` : answer + evidence[ ] + related_components[ ]
50- - ** Required fields** : Every component must include ` name ` , ` file_path ` , and optional ` line_number `
51+ - ** Required fields** : Core component arrays must include ` name ` , ` file_path ` , and ` line_number `
52+ - ** Flexible fields** : Complex nested types (coupling_metrics, dependencies, etc.) accept either structured objects or simplified strings
5153- ** Provider integration** :
5254 - Added ` response_format ` field to ` GenerationConfig `
5355 - OpenAI compatible providers send JSON schema to LLM API
56+ - Ollama-native ` format ` field for compatibility (dual-field approach)
5457 - AutoAgents adapter converts ` StructuredOutputFormat ` to CodeGraph ` ResponseFormat `
58+ - xAI/Grok supports Responses API with full ` response_format.json_schema ` support
59+ - ** Prompt updates** : All 28 tier prompts (4 tiers × 7 tools) updated to use structured format
5560- ** Hybrid output** : Combines freeform ` analysis ` field with structured arrays
5661- ** MCP handler** : Parses structured JSON and surfaces in ` structured_output ` field
5762- ** Benefits** :
58- - File paths are ** mandatory** - no more abstract references
63+ - File paths are ** mandatory** in core component arrays
5964 - Downstream tools can navigate directly to relevant code
6065 - Consistent data structure for programmatic consumption
6166 - Better agent-to-agent collaboration with actionable locations
67+ - Flexible parsing handles LLM variations while preserving essential data
6268
6369#### ** File Location Requirements in Agent Outputs (Deprecated)**
6470- ** Superseded by** : Structured output enforcement with JSON schemas (above)
0 commit comments