Core Concepts

Build sequences

How releases, dependencies, and priority/effort combine to decide what to build next.

A story map isn't just a flat list — it has enough structure to answer "what should I build next?" without a human deciding it every time.

Releases

Stories belong to a release. Release 1 is the MVP — in INTENT.md it's the ## MVP stories — build these first section; later scope lives under ## Release 2 — ... and so on. Releases are a hard scope boundary: an agent asked to build release 1 shouldn't reach into release 2's stories.

Dependencies

A story can depend on other stories. A story isn't eligible to be picked up until every story it depends on is done.

"What's next" resolution

This is exactly the logic behind intent next (CLI) and the equivalent MCP flow:

  1. If a story is already in_progress, surface it and stop — finish what's in flight first (skip this with --skip-wip on the CLI, or by working a specific story directly via mark_in_progress/start_story).
  2. Otherwise, filter to todo stories whose dependencies are all done.
  3. Sort by: build sequencereleasepriority (must before should before could) → effort (S before M before L before XL).
  4. Take the first one.

If every remaining todo story has unfinished dependencies, there's no valid next story — the tooling surfaces that as a stop, not a guess. If everything is done, that's a completion, not a next story.

Release readiness

Before starting a release, it's worth checking that every story in it is actually buildable — has acceptance criteria, has an implementation plan where one's expected. The remote MCP server's check_release_readiness tool does exactly this: a pre-flight scan across every story in a release that reports gaps (missing ACs, missing implementation plans, missing test criteria) before an agent starts building blind. See MCP → Tools & resources.

  • CLI: intent next computes and shows the single next story, one at a time.
  • MCP: the get_build_order tool (remote server) or the intentdocs://project/build-order resource (local server) returns the full recommended sequence at once, useful when an agent wants to plan several stories ahead rather than asking one at a time.