version: "1.0.0" name: aegis-context description: AEGIS architecture and source navigation. Use when implementing, debugging, testing, or explaining AEGIS behavior, or checking technical documentation against source.

AEGIS source navigation

Start with docs/development/current-context.md in the selected checkout. It owns checkout selection, current entrypoints, evidence limits and history routing. If working on an older revision without it, read that checkout's AGENTS.md, package.json, Vite and TypeScript configs and record HEAD and dirty state. Resolve source paths against that checkout, never the skill installation folder.

For a source question, follow imports and actual callers from the configured entrypoint. In main, inspect platform/index.js for OS dispatch and preload.js with the handler and renderer consumer for IPC. Use rule-loader.js and its callers to identify rule consumers. Read only the affected historical entries through memory-bank/README.md; preserve dated receipts and active parallel work.

Derive maintained counts with npm run counts:check and versions from package files. Check effective checkJs/include and per-file directives before describing type coverage. For generated inventory scope and optional dependency graphs, read the relevant sections of docs/development/workflow.md. Direct test imports and static graphs locate candidates; they do not establish behavioral coverage or complete runtime reachability.

When Serena is connected, activate the exact checkout by absolute path and verify its active project before symbol queries. Use symbol definitions/references for large files and literal search for IPC strings and computed CommonJS paths. Svelte mode includes JS/TS. A configured server is not proof of a connection.

Use design-system for interface work, electron-main for main-process changes, svelte-patterns for Svelte and testing for harness decisions. Commit/PR workflows remain separate and follow the requested scope. For skill-routing evaluation, read trigger cases; they are examples, not tasks or evidence that an agent evaluation ran.