Semantic Search MCP Server

{"id": "infinigen_terrain_camera_visibility_pipeline", "description": "Multi-file pipeline connecting terrain mesh generation, BVH tree construction, camera frustum culling and vertex visibility raycasting", "repo_path": "infinigen", "query": "I need to understand how Infinigen determines which terrain vertices are visible from the camera. The system seems to use BVH trees and raycasting, but I can't figure out how the terrain mesh, camera poses, and visibility checks connect across the codebase. Where does the full visibility determination pipeline live?", "difficulty": "hard", "expected_items": [{"file_path": "infinigen/terrain/utils/mesh.py", "must_include_substrings": ["def camera_annotation(self, cameras, fs, fe", "vertex_attributes[\"invisible\"]", "np.linalg.inv(cam_pose)"]}, {"file_path": "infinigen/core/placement/camera.py", "must_include_substrings": ["terrain_camera_query", "terrain.compute_camera_space_sdf", "placeholders_kd.find(cam.matrix_world.translation)"]}, {"file_path": "infinigen/assets/objects/trees/utils/mesh.py", "must_include_substrings": ["def get_visible_vertices(cam, vertices", "C.scene.ray_cast(", "is_visible[i] = False"]}, {"file_path": "infinigen_examples/generate_nature.py", "must_include_substrings": ["terrain_inview_bvh = mathutils.bvhtree.BVHTree.FromObject(terrain_inview, deps)", "split_in_view.split_inview("]}]} {"id": "codeql_promise_taint_propagation_async_await", "description": "How taint flows through Promise chains including then/catch/finally callbacks and async/await unwrapping across multiple library files", "repo_path": "codeql", "query": "I'm trying to understand how CodeQL tracks tainted data through complex Promise patterns. When tainted data enters a Promise.then() callback, then gets returned from an async function, then awaited elsewhere - how does the dataflow analysis connect all these steps? I need to find the core taint propagation logic for Promises.", "difficulty": "hard", "expected_items": [{"file_path": "javascript/ql/lib/semmle/javascript/Promises.qll", "must_include_substrings": ["module PromiseFlow", "predicate loadStep(DataFlow::Node promise, DataFlow::Node value, string prop)", "AwaitExpr await | await.getOperand() = promise.asExpr()", "call.getMethodName() = \"then\" and promise = call.getReceiver()"]}, {"file_path": "javascript/ql/lib/semmle/javascript/Promises.qll", "must_include_substrings": ["class PromiseTaintStep extends TaintTracking::LegacyTaintStep", "override predicate promiseStep(DataFlow::Node pred, DataFlow::Node succ)", "pred = thn.getCallback([0 .. 1]).getAReturn()"]}, {"file_path": "javascript/ql/test/library-tests/InterProceduralFlow/async.js", "must_include_substrings": ["async function async()", "let sink2 = await async(); // NOT OK", "let sink9 = unpack(await (pack(source)))"]}]} {"id": "qgis_processing_parameter_validation_chain", "description": "The complete parameter validation chain from CLI input through Processing registry to algorithm execution including C++ and Python layers", "repo_path": "QGIS", "query": "When running a QGIS processing algorithm, parameter validation happens at multiple levels - the C++ core validates parameter definitions, Python wrappers check values, and the algorithm itself has custom validation. I need to trace the complete validation chain from user input to algorithm execution. Where are all the validation checkpoints?", "difficulty": "hard", "expected_items": [{"file_path": "src/core/processing/qgsprocessingalgorithm.cpp", "must_include_substrings": ["bool QgsProcessingAlgorithm::checkParameterValues", "def->checkValueIsAcceptable( parameters.value( def->name() )", "invalidSourceError( parameters, def->name() )"]}, {"file_path": "python/plugins/processing/core/Processing.py", "must_include_substrings": ["ok, msg = alg.checkParameterValues(parameters, context)", "alg.validateInputCrs(parameters, context)"]}, {"file_path": "src/process/qgsprocess.cpp", "must_include_substrings": ["p->checkValueIsAcceptable( params.value( p->name() )", "alg->checkParameterValues( params, context, &message )"]}]} {"id": "claude_flow_swarm_failure_recovery_distributed", "description": "Agent failure detection, task reassignment, and recovery coordination across SwarmCoordinator, AgentRegistry, and TaskCoordinator", "repo_path": "claude-flow", "query": "When an agent fails during task execution in Claude Flow's swarm system, there's a complex recovery mechanism. I need to understand how failure is detected, how tasks get reassigned to other agents, and how the system maintains consistency across the SwarmCoordinator, AgentRegistry, and TaskCoordinator. Where is all this failure handling logic?", "difficulty": "hard", "expected_items": [{"file_path": "src/swarm/coordinator.ts", "must_include_substrings": ["async failTask(taskId: string, error: any)", "isRecoverableError(error)", "isRetryableError(error)", "agent.errorHistory.push", "task.attempts.length < (task.constraints.maxRetries"]}, {"file_path": "src/hive-mind/core/Agent.ts", "must_include_substrings": ["protected async handleTaskFailure(taskId: string, error: any)", "await this.sendMessage(null, 'task_failed'", "this.emit('taskFailed', { taskId, error })"]}, {"file_path": "src/core/AgentRegistry.ts", "must_include_substrings": ["async coordinate(task: Task): Promise<Result>", "findSuitableAgents(task)", "selectBestAgent(suitableAgents, task)", "updateAgentPerformance"]}, {"file_path": "src/task/coordination.ts", "must_include_substrings": ["async coordinateSwarm(", "coordinateCentralizedSwarm", "coordinateDistributedSwarm"]}]} {"id": "infinigen_constraint_solver_furniture_placement", "description": "How the constraint language evaluator connects to simulated annealing solver for furniture placement optimization", "repo_path": "infinigen", "query": "Infinigen uses a custom constraint language to define furniture placement rules, but I can't figure out how these high-level constraints get converted into an optimization problem that the simulated annealing solver can work with. Where does the constraint evaluation and solver integration happen?", "difficulty": "hard", "expected_items": [{"file_path": "infinigen_examples/constraints/home.py", "must_include_substrings": ["def home_furniture_constraints():", "cl.Problem(", "constraints=constraints", "score_terms=score_terms"]}, {"file_path": "infinigen_examples/configs_indoor/base_indoors.gin", "must_include_substrings": ["SimulatedAnnealingSolver.initial_temp", "SimulatedAnnealingSolver.final_temp", "compose_indoors.solve_steps_large"]}, {"file_path": "infinigen/core/constraints/example_solver/annealing.py", "must_include_substrings": ["class SimulatedAnnealingSolver:", "def step(self, consgraph, state, move_gen_func, filter_domain):", "metrop_hastings_with_viol"]}]} {"id": "codeql_jsx_parse_fallback_chain", "description": "The complete JSX parsing fallback chain from file extension detection through multiple parser configurations to final AST generation", "repo_path": "codeql", "query": "CodeQL's JavaScript extractor has a complex fallback mechanism for parsing JSX files. It first tries based on file extension, then falls back to enabling various parser options. I need to understand the complete chain of parser configuration attempts and how each fallback is triggered.", "difficulty": "hard", "expected_items": [{"file_path": "javascript/extractor/src/com/semmle/js/extractor/JSExtractor.java", "must_include_substrings": ["textualExtractor.getLocationManager().getSourceFileExtension()", "JSParser.parse(config, sourceType, extension, source", "ES2015Detector.looksLikeES2015(parserRes.getAST())", "if (wrongGuess)"]}, {"file_path": "javascript/extractor/src/com/semmle/js/parser/JcornWrapper.java", "must_include_substrings": ["public static boolean alwaysParseWithJsx(String extension)", "return extension.equals(\".jsx\")", "if (alwaysParseWithJsx(extension)) options = new JSXOptions(options)", "catch (SyntaxError e)", "if (config.isJsx()) options = new JSXOptions(options)"]}]}