What problem does it solve? AI agents work better with a small set of verified, non-derivable project truths than with generated documentation. This Skill builds and maintains that context system: a tiny always-loaded kernel file plus a bundle of small verified knowledge entries covering architecture rationale, conventions, landmines, and org requirements. ## Core Features & Use Cases - Ingest, Query, and Audit intents: Build context from brownfield codebases or greenfield specs, answer questions from the bundle without loading it all, and shrink or re-verify stale entries over time. - Deterministic mechanics script: A bundled Python CLI handles config resolution, index generation, validation, staleness sweeps, cross-project resolution, and syncing kernel content into AGENTS.md files. - Trust tracking and memlog: Every claim is marked verified or generated, every decision lands in an append-only memlog, and refresh runs diff against prior state instead of starting over. - Use Case: Point it at a legacy repo and say "generate project context" — it scans code and docs, confirms claims with you in short rounds, and writes a kernel.md with commands, conventions, and landmines plus linked bundle entries. ## Quick Start Ask the agent to generate project context for this repository and answer its confirmation questions as it scans the code and docs.
