Live Runs — the long-horizon prompts, executed
Every prompt in the Kit was written to be run, so this page runs them: the same briefs, count contracts, verification gates and verifier verdicts the prompts mandate — with the artifacts they produced linked below. Nothing here is a demo of a run; it is the run log.
Status
| Run | Prompt(s) | Brief | State |
|---|---|---|---|
| A — Prompt Composer | 01 Single-Agent + Brief Template | Brief A | executing — plan 3 of 7 done |
| B — Launch Content Kit | 02 Orchestrator + 03 Worker Spec + 04 Verifier + Brief Template | Brief B | brief written; workers pending |
Run A — plan (frozen under Prompt 01's count contract)
- Write
runs-src/composer.html— single-file interactive composer (inline CSS/JS, no external deps, 13 categories, role picker, 3 toggles, live output, Copy) — done node --checkon extracted inline JS → exit 0 — done (exit 0)- Strict HTML tag-balance parse → exit 0 — done (exit 0; first run of the checker itself crashed on a missing import and was fixed — the checker, not the page)
- Wire into
build-site.sh;./build-site.shexits 0 with page count ≥ 16 — in progress - Deploy;
curlreturns 200, size > 10,000 bytes — pending - Behavioral check in a real browser (fill form → Assemble → assert all 4 ARC(E) sections + role string + QA text) — pending
- Evidence audit; state "7 of 7 done" — pending
Prompt 01's rule in force: partial work does not count, a plan does not count, and "done" requires pasting the final passing output.
Artifacts
- Prompt Composer — interactive ARC(E) prompt builder (Run A output)
- Brief A — the single-agent brief, success predicate first
- Brief B — the orchestrator brief for the launch kit
(Run B artifacts — campaign brief, email sequence, social calendar, ad set — and the fresh-context verifier's verdicts land here when the orchestrator round completes.)