Diagnose local agent and project setup problems — missing commands, broken MCP configs, stale generated assets, bad paths, invalid catalog entries, and missing env vars.
Inspect changed dbt assets, estimate blast radius, identify missing tests, and recommend the narrowest safe validation plan.
Use to enforce consistent naming conventions and file structures across a project without changing business logic.
Compact context at logical phase boundaries — after research, after planning, after debugging — rather than mid-task. Preserves useful state while clearing noise.
Use when designing or using MCP-backed structured code search with search, AST query, symbol inventory, and bounded extraction workflows.
Use when about to commit, push, or publish -- staged changes touch config or environment files, generated artifacts (relay sessions, logs, build output) are being added, or the…
Write state-restoration documents for passing tasks between agents or engineers.
Relay questions and tasks to a persistent local Gemini ACP session with cross-turn context.
Use when a graph already exists and the user needs retrieval, tracing, explanation, or gap detection from graph memory before reopening the full corpus.
Audit whether a dbt incremental model uses the right incremental strategy for the repo, the data shape, and the operational constraints.
Analyzes ML training scripts to enforce seed setting, deterministic operations, and environment tracking for exact reproducibility.
Use when the user provides a MarkItDown-supported file or URL and converting it to Markdown first will make inspection easier, cheaper, or more reliable.
Use when querying, ingesting, or maintaining a local RAG MCP corpus for semantic document retrieval with privacy controls.
Ultra-compressed communication mode. Cuts token usage ~75% by dropping filler, articles, and pleasantries while keeping full technical accuracy.
Relay questions and tasks to a local Codex CLI using one-shot subprocesses per ask.
Standardize the reporting of model metrics to ensure statistical rigor and business relevance.
Relay questions and tasks to a local Claude CLI using one-shot subprocesses per ask.
Multi-model orchestration — route to two external advisors, then synthesize
Use when code needs restructuring without changing observable behavior.
Process-first advisor routing with artifact capture
Ensure that new complex models actually outperform simple, naive baselines.
Extract a learned skill from the current conversation manually
Identify and prevent target leakage in ML preprocessing pipelines.
Audit the methodology used to split data into train, validation, and test sets.
Turn vague implementation requests into testable requirements before coding.
Verify that API endpoints match their OpenAPI/Swagger specifications.
Use to diagnose tests that pass and fail intermittently without code changes.