2
3 Comments

Obserf 0.3

The review inbox got a real redesign

Quick follow-up to last week's post. Turns out the part of Obserf I use most isn't the scanning, it's sitting down with 30 findings and deciding what's worth a reply. That inbox was a pile of buttons, so 0.3 is mostly a redesign of it.

What changed:

- Each finding shows the model's verdict first, then the actual page text, then the score. "Why 55?" shows the math, because a number without its reasoning is just vibes with a decimal point.

- Everything has a keyboard shortcut. Press ? to see them all. I can get through a queue without touching the mouse now.

- Undo sits right where you made the decision, because d is one key away from s.

- If a scan finishes while you're reviewing, you get a "refresh?" banner instead of the list reshuffling under your cursor.

- Scores are plain numbers now. The green/amber colours implied a precision the ranking hasn't earned yet.

Still never posts anything. Still local, still free, still just Bun + Claude Code.

Upgrade: bun add @obserf/cli@0.3.0

New: bunx @obserf/cli init ~/dev/marketing

Honest question for anyone who's tried it: does the ranking hold up on your projects, or does the top 10 still need a lot of squinting?

https://github.com/kriasoft/obserf/releases/tag/v0.3.0

posted toAvatar for product Obserf
Obserf
  1. 2
    Have early users shown that the revised ranking consistently surfaces the findings they actually choose to act on, or is the main improvement still the review workflow?
    1. 1
      No, the ranking itself didn't change in v0.3. score() returns the same numbers; I only added a function that shows the math behind a score. This release is all review workflow. The groundwork for checking the ranking went in with 0.2. Each scan freezes its top 10, and I mark each finding as shortlisted, skipped ("good, not for me") or dismissed ("shouldn't have shown me this", with a reason). My bar is 5/10 worth a look and nothing embarrassing in the top 3. Honestly, so far that's mostly my own projects, so I can't call it consistent yet. Part of why I redid the inbox was to make labelling fast enough that I'd keep doing it. If you try it, your dismissal reasons would help me more than anything else.
      1. 2
        Could be useful to compare notes once you have more of those labels from other users — happy to continue by email sometime.