gyst

Tests Gyst compatibility with self-hosted LLMs running through Ollama, vLLM, and OpenCode. Use when the user mentions "self-hosted," "local LLM," "Ollama," "vLLM," "OpenCode," "private," or "on-premise." Also use when testing MCP compatibility across different clients.

Analyzes current algorithms in Gyst and proposes specific improvements based on real eval data. Use when the user says "improve," "optimize," "better search," "fix retrieval," "algorithm," "brainstorm," or when eval metrics show degradation. Also use after running retrieval-eval.ts.

Enforces code uniformity across the entire Gyst codebase. This skill activates on EVERY file edit in src/. It checks patterns, naming, imports, error handling, logging, and database access. Use when writing, reviewing, or modifying any TypeScript file. Auto-activate always.

Sets up and maintains CI/CD pipelines for the Gyst project using GitHub Actions. Use when the user mentions "CI," "CD," "pipeline," "deploy," "actions," "workflow," "release," or "publish." Also use when creating or modifying .github/workflows/ files.

Detects bugs, logic errors, and unsafe patterns in TypeScript code. Use when writing new functions, reviewing code, fixing errors, or when the user says "check for bugs" or "review this." Auto-activate when editing files in src/.

Enforces consistent code patterns across the Gyst codebase. Use when writing new code, reviewing changes, or when the user mentions "consistency," "uniformity," "patterns," or "refactor." Auto-activate on any src/**/*.ts file.

Collects git logs, error messages, conventions, and project metadata from local repositories to build Gyst evaluation datasets. Use when the user says "collect data," "build test fixtures," "get git logs," "seed the database," "gather project data," or "build eval dataset." Also use when setting up the retrieval evaluation harness.

Optimizes SQLite queries, schema design, and FTS5 usage for the Gyst knowledge base. Use when writing database queries, modifying schema, debugging slow searches, or when the user mentions "slow," "performance," "query," "index," or "database."

Identifies edge cases and boundary conditions that need handling. Use when implementing new features, writing tests, or when the user says "what could go wrong" or "edge cases." Activate when creating or modifying files in src/.