Lab context
Instructions
Goal
Run the local verification command and capture the exact evidence that proves the branch is still healthy.
Hard gates
- Use the house path from this lab route.
- Run work through the agent, not browser editing.
- Keep evidence tied to the frozen checkpoint.
- Finish through evaluation-backed progress from this lab route.
lab
Run the first verify gate
Ready when you are
Lab completed.
The released path already records this lab as complete. The reference bundle remains available for review; selected run evidence and evaluation stay attached to this page.
Temporary seam: Start and continue post through this lab route, use BFF-backed Agent Runner/Evaluation/Learning Progress contracts, materialize the fake Sandbox, and return read-only run and result evidence. Later slices add abort, failure panels, quota/paywall, hosted preview, and real checkpoint runtime states inside the lab page. Still not verified: Real SB08 checkpoint runtime, durable artifacts, quota/paywall gates, and full fail-loop handling remain outside WS06C.
fresh checkpoint · no leftoversRuntime
BFF-backed Agent Runnerfake-fast model · 512 token cap · no silent provider switchSandbox checkpoint
orientation/first-verify-run2026-05-30.catalog-seed.v1 · frozenCheckpoint registry APIHeads up
No browser editingThat is not a missing feature. It is the lab constraint.What will happen
- Start lab actionThe lab route selects the Catalog task and methodology pins.
- Materialize sandboxThe runner uses the frozen checkpoint metadata.
- Agent works, logs streamYou inspect evidence. You do not hand-edit code.
- Evaluation closes the loopThe lab route records evaluation-backed progress and exposes the next house step.
Reference bundleFirst verify run reference bundleOpen spoiler bundle
Golden diff, ideal AGENTS.md, review notes, and common failure examples for the first verify lab.
- Golden diff
- Ideal AGENTS.md
- Verify output
- 3 common failures