USER PILOT · REVIEW ON 23 OCT 2026
Does it earn
a second use?
Help decide whether Second Look is worth building further. Try one real task. Tell us what you adopted, rejected, or found too hard to prepare.
The preview uses hand-authored candidates. No installation or login needed to view it.
A SMALL, REAL TASK
Is this for you?
You have an existing static HTML/CSS/JS page with something unfinished, or an AI revision you want to check.
You can run Python 3.11+ on macOS or Linux. If you need a new candidate, use the AI tool you already have.
Backends, arbitrary repositories and Python execution are outside this beta. Don't buy a subscription or API credits for the pilot. Your existing tool's normal quota and generation cost still apply.
VALUE BEFORE MORE FEATURES
What we're testing
Can 3 external developers complete a real comparison? Do 2 return for a different task?
We also need an adopted improvement or avoided regression, and measured time or token benefit over the usual method.
Features are on hold while we learn. Stars, downloads and an intention to return don't count as actual use. Results are pending.
ONE ATTEMPT IS USEFUL FEEDBACK
Bring work you already care about.
- Pick a small task. Write what it should do, one improvement to make and one behavior to preserve. Keep those criteria fixed.
- Start a timer. Install v0.6, then follow the own-project capture guide. Stop and report the barrier if preparation exceeds 20 minutes.
- Compare a candidate. Follow prepare → external generation → assess. The external generation and saving its response are manual. Local assessment makes no model calls.
- Make an actual decision. Inspect checks, regressions and screenshots. Keep the original, adopt a change, or reject the comparison. Tell us why.
Twenty minutes is a preparation limit, not a promised completion time. Include installation and writing checks. Record generation/waiting and review time separately.
GitHub sign-in required to submit. Issues are public. A short description and timings are enough; keep private code, prompts, logs and screenshots out. Unknown usage is fine.
Did another real task bring you back?
Add a second-task comment to your first issue. A rerun or another candidate for the same task doesn't count as a return. Don't create a task just to meet the pilot's target.
Optional: compare with your normal workflow
Use the same starting artifact, goal, model/settings and acceptance criteria in fresh copies. Compare a normal direct request plus manual review with Second Look. Record method order, all preparation/review time, waiting, failures and comparable usage if available. Stop within your existing quota.
Decision checkpoint: 23 October 2026, KST. A small pilot can justify another experiment; it cannot establish SOTA or general market demand.
Read the fixed decision rules ↗