product-demo-seeding

Build from a clean detached worktree of freshly fetched origin/main, then create fictional workspaces, tasks, sessions, and provider state through the E2E fixture and ApiClient. Seed only enough coherent data to make the story legible, rehearse without a recorder, and reseed a pristine baseline per take.

Run deterministic mock GitHub and Jira providers at the backend boundary with fixed IDs, timestamps, actors, and ordering from one versioned fixture document. Keep provider credentials unset, block live network calls, and prove mock routing in backend logs before recording.

No. The skill forbids copying or querying a developer's database or instance, including scripts/dev-isolated --copy-db. Every take starts from a fresh disposable database with its own temp HOME, ports, display, and browser profile.

Rehearsal and failed takes mutate product state, so recording against the same database accumulates duplicate tasks and provider-created records. Start each recorded take from a fresh database or a proven complete reset, and assert exact record counts before RECORD.

DOM deletion, CSS hiding, and capture-time text replacement are prohibited because they mask real product state and responsive defects. If a surface is broken, report the bug and choose another truthful native route that demonstrates the same capability.