End-to-end orchestrated execution. Takes a feature description and drives the entire pipeline — plan, plan review (3 fresh reviewers), WU decomposition, 4-phase execution loop per WU, final review, COMMIT-READY emit. Honors user's git policy (no auto-commit).
`--default-domain` comes from the project-detect framework signal (orchestrator Step 5b / project-context); it only decides how UI files with no strong path marker are classified. `--risk-tier` is the WU's `risk_tier` fi
**Capture the WU baseline ONCE per WU, before attempt 1 (NOT on retries).** This is what makes file-scope checks correct across a multi-WU run where the user has not yet committed prior WUs (per the `ben yapacagim` git p
This gate has **two interchangeable execution paths that MUST emit the identical aggregated verdict object** (Step 3). The prose path below is the canonical fallback and the single source of truth for reviewer prompts an
A baton-based system for building mobile features across multiple Claude Code sessions with persistent progress tracking and simulator verification between steps.
Set up end-to-end testing framework with CI integration
Scaffold a Supabase/Playwright E2E harness for a Refine/TanStack web app using Telar's Arrange/Act patterns
The most common mistake in Supabase E2E suites is letting `service_role` do double duty: it provisions the fixtures AND performs the action under test, which means the test never actually proves RLS works — it just prove