Skip to content

Release v4.0.0

Latest
Latest

Choose a tag to compare

View all tags
@github-actions github-actions released this 26 Oct 22:43
v4.0.0
e803d51

🚀 The Agentic Startup v4.0.0

Highlights

  • One spec file: requirements, solution and plan live together in spec.md.
  • Fewer interruptions: questions come in batches, and only about behavior your users will notice.
  • Specs stay within your brief: no quality targets, data rules or design choices you didn't ask for.
  • Faster builds: each increment is built and checked as a whole, and only failed checks are re-run.
  • Lighter test setup: end-to-end tests are added only if you want them.

Added

  • Spec event log: spec.js and task.js append every gate, requirement, decision, question, assumption, task change, run and result to the spec's events.jsonl. spec.js <id> --event records an event; spec.js <id> --snapshot starts a log for an existing spec.
  • Dashboard mod: a Claude Code mod that shows the spec's gates, requirements, phases, slices and scenarios in a side pane while a spec skill runs, replayed from its event log. /start-dashboard [spec-id] opens or closes it.
  • End-to-end tests on request: when a project has no end-to-end setup, the plan asks once whether to add one. The suite runs once after every slice is built, not in each TDD step.

Changed

  • BREAKING: one spec.md per spec, with Requirements, Solution, Plan and Ledger parts, replaces requirements.md, solution.md and plan.md. /start:implement still reads the plan.md of an older spec.
  • BREAKING: /start:implement builds and evaluates per phase. Code agents build all of a phase's slices, a service agent restarts the app, and one eval agent scores the phase with a holdout test script kept outside the project. Retries re-run only failing scenarios; passed and building slices are skipped.
  • Plans have fewer, larger phases, each ending in something a user can see working; a small spec is one phase. Every slice carries its features through each interface the spec defines, and the plan shows which slices run in parallel and which wait on another.
  • /start:specify asks fewer questions: up to four per call, and only for product rules users could tell apart. Other defaults are confirmed with each part's approval, and a plan of one or two phases gets a single approval.
  • /start:specify leaves UX, quality and data choices to the brief: the UX, Security and Performance research lenses and design guidance are gone, and quality gates and data obligations are recorded only for concerns the brief or user raises.
  • Code agents and the stages that judge them run on the session model.
  • /start:implement lists the recorded gaps of built slices without offering a re-run, and names the next commands instead of asking about git or archiving.
  • The test skill no longer names specific browser runners.

Removed

  • The separate requirements, solution and plan templates, the spec report template, the example PRD and architecture examples, and the solution's User interface section.

Installation

Via Claude Code Marketplace (Recommended)

/plugin marketplace add rsmdt/the-startup
/plugin install start@the-startup
/plugin install team@the-startup   # optional

Manual Installation

Download the archive for your platform:

  • the-startup-v4.0.0.tar.gz — Linux/macOS
  • the-startup-v4.0.0.zip — Windows

SHA256 checksums are available in checksums.txt.

Requirements: Claude Code v2.0+ with marketplace support.

Assets 5
Loading