Skip to content

v1.12.0

Choose a tag to compare

@phuryn phuryn released this 14 Nov 22:39
· 8 commits to main since this release

Every stage now ends with end-to-end tests of what it changed, real sign-in tested, and current docs.

Why: a product built with these skills ran two weeks of stages, 77 review rounds and nine development deploys without one browser test of the running app. Its gate never ran e2e or e2e:auth, every test mocked the session, and its core docs were written only after the work was lost. The skills described the browser suite, but owed it only before a production promotion, and let a plan leave its stage open behind later ones.

What changes for you, by default (your plan can choose otherwise; the choice is written down):

  • The full gate runs at each stage's end and before each deploying push, not only before production, and it holds every suite in the contract your product's shape has (test-harness). Compare your gate script with it: a green script can be missing a whole suite.
  • Each stage ends with committed end-to-end tests of the journeys it added or changed. Checking a flow means writing its test once; a scratchpad script or a click-through doesn't count, and a test with a mocked session doesn't cover a signed-in journey.
  • The signed-in browser suite and the core docs come with the first real slice (app-lifecycle stage 5). Clerk's test mode, your own verification code and a first test account move to Setup (stage 4); stage 6 now proves the permission matrix in the browser. Stage numbers are unchanged.
  • Only the owner skips or defers a stage, recorded in the plan with what brings it back. A broad go waives no stage it doesn't name.
  • At each stage's end, document-app runs in check mode beside the review, and the stage's line in the plan is closed against its done-when or says what keeps it open. Docs say what is built now.
  • Previews and go messages name every suite that didn't run. Counts alone no longer pass for coverage.
  • Builders run the tests they write and fix them before the hand-off, and name any test they changed to make it pass. Deploys, applied migrations, pushes, provider settings, secrets and production credentials stay with the session agent. A Codex build that runs tests uses --sandbox danger-full-access by default, with its brief as the boundary: on macOS, Codex CLI 0.166.0, workspace-write could not start Chromium or resolve DNS (codex-agents, delegation).
  • derive-tests has an end-to-end test type and maps each journey in your flows doc to its spec; flight-deck keeps anything a later session needs out of scratchpads and system temp.

What to do: update the packs. If your product is past its first real slice without a signed-in browser suite, or its core docs, plan one catch-up stage for them; until then, every preview and go message will say which suites didn't run.

Skills changed: test-harness, app-lifecycle, release-gate, derive-tests, shipping-artifacts, delegation, codex-agents, claude-agents, grok-agents, gemini-agents, flight-deck, agent-instructions (its product-instructions skeleton), clerk-production, start-here. Reviewed in round 148 (fresh Codex session on GPT-6.1 Sol; Grok was out of usage): nothing above Low.