start-here finds where a product’s work stands and names the next step along its lifecycle, from idea to paying users and round again. It runs only when you start it by name: /product:start-here in Claude Code or $product:start-here in Codex.
It checks whose repository the folder is, reads the repository’s instructions, docs and history, fetches what the remote has, and then points to the skill that does the next step. A small fix goes straight to the build, with no proposal or plan.
What You Get
- A named next step and the skill that handles it
- A plan prompt for complex changes that opens with the agent’s key assumptions
- A one-time pass wiring installed packs into the product’s agent instructions
- A walk through the eight lifecycle steps from proposal to real users
Common Use Cases
- Starting a new product from an idea with no repository yet
- Resuming work and finding where it stands
- Deciding how to handle a bug, a small change or a complex change
- Preparing a release with the right checks
What Makes It Unique
The skill is the entry point of the whole product pack and runs only by name, so it never fires on its own. It checks that a repository belongs to the product before reading it, fetches the remote before trusting local docs, and keeps small fixes out of the proposal and plan steps. It also gives the owner a fixed plan prompt that moves decisions needing a person to the top.
Under The Hood
The skill is one SKILL.md of instructions with a small agent configuration file and a receipts.md of rule origins. It tells the agent to run git commands: rev-parse to find the repository, fetch, log and a fast-forward pull when the tree is clean, remote set-head, and git init in an empty folder or with the owner’s word. It reads the repository’s agent instructions, docs, plan status and history, and records the owner’s answer about a repository in the agent instructions for the next commit. It runs no scripts of its own and does not call web services.