Write the agent-skill for the product you just built, so a coding agent can drive it. Reads the product's own API surface — Next.js route handlers, an OpenAPI spec, or an eve agent's tools — and produces a distributable skills/<slug>/SKILL.md in the product repo: auth (device flow before raw keys), the loop an agent should perform, the endpoint reference, error shapes, and a common-mistakes table grounded in what actually breaks. Plus the README row and the npx skills add owner/repo line for the product's users. Use when the product exposes an API that is not only for its own frontend, when the user says "make my product usable from Claude Code / Cursor", "ship an agent skill for this", "how do agents use my API", or when dev-flow raises it near feature_complete. Refuses when the product has no callable surface. Not for: skills that operate ON this repo (those are dev-flow's own), and not a docs generator — the output is a runbook, not a reference dump.
dev-flow raises it at feature_complete → deployed, once there is a/plugin install dev-flow@dev-flow # the whole suite
./install.sh --platform claude # or drop dist/product-to-agent-skill.skill into Claude Code