orc-doc

What problem does it solve? Writing a long document such as a PRD, TSD, or runbook inside a single AI session exhausts context, loses progress when a session dies, and cannot be resumed later. This Skill writes long documents section-by-section with each section stored as its own file, so work survives session limits and resumes months later without re-explaining anything. ## Core Features & Use Cases - Section-based writing pipeline: Each section lives in its own file under sections/ as the source of truth, and document.md is rebuilt from them for free with orc doc compile. - Bounded agent dispatch: The orchestrator never reads the document body; writers each own exactly one file and checkers each read one bounded part, keeping context usage small even for 10,000-line documents. - Resumable waves with cost control: Every wave is a stop, partial mode writes one wave before you pay for the rest, and RESUME.md lets a fresh session continue exactly where it stopped. - Free deterministic linting: Rules like no questions in the body, N/A instead of filler, and no local-only references run without model tokens before any paid check. - Use Case: Ask it to write a PRD for a checkout refund feature; it interviews you once, freezes the context to disk, writes the document in waves you can review, and hands you a Markdown file that imports cleanly into Notion, Google Docs, or GitHub. ## Quick Start Ask the assistant to write a PRD for your feature using /orc-doc and answer the short intake questions about purpose, audience, and template.