The failure this exists to catch is a session that reached green by deleting an assertion, skipping the flaky test, or quietly shrinking the task down to the part that passed. None of that ever leads a summary. Codex is comfortable reading session data and diffs closely enough to find the four line change sitting behind a confident claim, and it quotes the session's own words when it reports one.
Run a staged conversation — no account needed. The agent handles it for real while a simulated world answers its tool calls; nothing touches real accounts, and nothing is actually sent.
Migrate the notifications service off Moment.js to date-fns
Ran 47 minutes. Opened pull request 318 in fleetwing/notifications across 61 files, CI green. The session summary calls the migration complete, while a later message notes one helper, the business day offset used by the digest scheduler, still calls Moment because date-fns carries no direct equivalent. The run ended with the pull request left as a draft.
Using this template drops you into a guided setup. It asks exactly this, nothing else:
Connect Devin
One sign-in. The agent acts through your account, scoped to what this template uses.
Connect Slack
One sign-in. The agent acts through your account, scoped to what this template uses.
Review rules
What finished actually means to you: what a session has to produce before you would call it shipped, the shortcuts you refuse to accept as a fix, whether the agent may message a stalled session back, and who owns the queue when one goes wrong.
Runs on Codex
Preselected for this page — connect your Codex account during setup, or switch to NoClick's built-in models with one click.
Watch it handle a test run
A staged conversation against a simulated world — then it’s live.
A test annotated skip with the note flaky, tracked separately, is a fix on paper only. Catching it means holding what was asked next to what the artifacts contain, line for line.
Every verdict here rests on something retrievable: a link, a file count, a prior session against the same task. That is well scoped work rather than open ended judgement.
Each finished session is its own run and its own Slack message, so two at once produce two notes rather than one confused one, and they can land out of order. The only thing ever written back is the single reply to a session waiting on input, and that reply is addressed to one session.
Free to start. Guided setup, a test run against staged conversations, and it's live.