What problem does it solve? Working on lci's code_insight MCP tool means navigating a 2400-line handler, a dozen analyzers under src/analysis, attribute-scoping gates, and golden tests, with no CLI or HTTP route to fall back on. This Skill maps every surface, code path, config knob, and known trap so you can debug artifacts, re-pin goldens, or benchmark the report without rediscovering the architecture. ## Core Features & Use Cases - Code map and surface table: Entry points, handler dispatch branches, LCF section emitters, and analyzer files (health, layers, modules, features, coupling, naming, clones, dead code, security, impact, error-handling BETA) with exact file:line references. - Probe recipes and test commands: Ready-to-run one-shot MCP stdio calls against the synthetic multi-lang corpus, plus gtest filters and ctest commands for goldens and real-project suites. - Invariants and traps: Documented pitfalls such as call-edge precision limits, dynamic-dispatch over-counting, attribute scoping (D1 fix), hash-order determinism, and stale persistent index servers. - Product comparison: Claim-by-claim comparison axes against SonarQube, CodeScene, CodeQL, lizard, madge, and fallow, with honest gaps (no dataflow, no persisted reports, MCP-only LCF output). - Use Case: You see an inflated cycle count under CYCLES for a Kotlin repo. Use this Skill to trace the artifact to bare-name call resolution, read the quality-plan doc it cites, and verify the fix with the unified probe recipe. ## Quick Start Ask the agent to run the unified code_insight probe against the multi-lang test corpus and explain any surprising section in the output.
