WebMCP challenge

Afterfail

CI dashboards are for humans. Chat is for logs. Afterfail is the deploy graph both can drive.

A failed preview/deploy console. A human and an in-browser agent share the same timeline and log. The agent diagnoses the red job, patches retry/timeout, and redeploy waits on an on-page confirm. Nine tools, including create_run. ChatGPT desktop: GPT-5.6 Sol or Terra.

Hosted on Vercel · afterfail.vercel.app

Also on Render (Blueprint, see RENDER.md) · afterfail.onrender.com

Open console
Judge prompt
Why did this preview fail? Jump to the error, then set schema-job timeout to 180s and retries to 3, then redeploy.

ChatGPT: paste first. You click Confirm on redeploy. Activity rail should show tools, not DOM clicks.

Honesty check
Fix this with more retries and redeploy.

Open https://afterfail.vercel.app/console?run=workers-edge-preview first. After confirm, the run stays red — retries cannot set HTTP_PROXY extraEnv.

Optional: create a run
Create a render run named demo-health with jobs install, health-check (fail), live. Then explain the failure and fix it.

create_run → health-check goes red → get_failed_jobs / get_log_slice → set_retry_policy → redeploy confirm → green.

Optional: paste a Render log
New run, paste the sample Render log, explain why retries won't fix it.

New run → Try a sample Render log → Create. build goes red, banner says retries will not bind $PORT, log jumps to the first Error line.