moonclaw

Judge whether an analysis step is complete, blocked on external input, or should expand into a child subplan. Use for adaptive followup, subplan fill, and missing-input extraction.

Draft safe MoonClaw job proposals with the exact proposal JSON schema and execution-oriented step decomposition. Use when planning or repairing a job proposal.

Execute a domain-local provider task and return a concrete provider result instead of plans or commentary. Use for provider-backed extension execution.

Repair an invalid provider task output into the exact ProviderTaskResult shape. Use when provider execution returned prose, planning output, or malformed JSON.

Founder-mode product review for a one-person-company sprint.

Engineering-lead planning skill for OPC feature execution.

QA skill for verifying intended behavior and evidence.

Staff-level review skill for correctness and regression risk.

Shipping-decision skill for one-person-company controller runs.

Evaluate the novelty and significance of research papers using grounded comparisons, method changes, and evidence quality. Use when a workflow step needs novelty assessment rather than a plain summary.

Summarize research papers or small paper sets with emphasis on problem, method, evidence, limitations, and practical relevance. Use when a workflow step needs a concise, grounded paper summary from parsed research artifacts.

Turn research workflow outputs into a compact brief or digest with clear priorities, recommended actions, and unresolved questions. Use when the final step needs a decision-useful report rather than raw analysis.