# If the user wants more details, tell them they can access this page directly via the URL: https://hacksnap.live/story/let-your-ai-agents-paint-big-arrows-boxes-and-text-on-your-screen-50018817

# Show HN: Let your AI agents paint big arrows, boxes and text on your screen

361 points · 157 comments

[Full discussion](<https://news.ycombinator.com/item?id=50018817>)

[Read original](<https://github.com/franzenzenhofer/big-arrow-on-the-screen>)

Category: [Agents & Coding](<https://hacksnap.live/?category=agents-coding>)

## Skept-o-meter & Hotness

Skept\-o\-meter: High\. Estimated from 10 comments\.

1 comment for the summary\.

Peak rank: \#2

Time in Top 10: 24\.0 hours

Hacksnap ranks recent stories first, then orders each group by points\. Peak rank uses all retained observations\. Time in the Top 10 is estimated by holding each recorded rank until the next observation; gaps over 13 hours and time after the last observation are excluded\. Movement between observations is unknown\.

27 recorded rank observations from 2026\-10\-09T13:02:30\.35034\+00:00 to 2026\-10\-10T23:01:00\.95027\+00:00\.

Hotness — latest 27 recorded Hacksnap ranks:

2026\-10\-09T13:02:30\.35034\+00:00: rank \#7

2026\-10\-09T14:01:34\.215134\+00:00: rank \#6

2026\-10\-09T15:01:27\.564392\+00:00: rank \#4

2026\-10\-09T16:02:06\.561056\+00:00: rank \#4

2026\-10\-09T17:03:06\.673648\+00:00: rank \#5

2026\-10\-09T18:01:02\.851052\+00:00: rank \#4

2026\-10\-09T19:01:11\.250291\+00:00: rank \#3

2026\-10\-09T20:01:32\.91907\+00:00: rank \#3

2026\-10\-09T21:02:08\.376144\+00:00: rank \#2

2026\-10\-09T22:01:19\.253075\+00:00: rank \#2

2026\-10\-09T23:01:31\.990912\+00:00: rank \#2

2026\-10\-10T08:01:00\.593376\+00:00: rank \#2

2026\-10\-10T09:01:45\.527975\+00:00: rank \#2

2026\-10\-10T10:00:53\.799353\+00:00: rank \#4

2026\-10\-10T11:00:35\.65851\+00:00: rank \#4

2026\-10\-10T12:01:11\.169647\+00:00: rank \#4

2026\-10\-10T13:00:54\.008377\+00:00: rank \#87

2026\-10\-10T14:01:43\.208766\+00:00: rank \#89

2026\-10\-10T15:01:24\.825188\+00:00: rank \#89

2026\-10\-10T16:01:10\.213328\+00:00: rank \#88

2026\-10\-10T17:00:48\.84189\+00:00: rank \#87

2026\-10\-10T18:00:57\.55132\+00:00: rank \#87

2026\-10\-10T19:00:52\.379463\+00:00: rank \#86

2026\-10\-10T20:00:21\.993912\+00:00: rank \#86

2026\-10\-10T21:01:10\.657738\+00:00: rank \#87

2026\-10\-10T22:00:50\.433569\+00:00: rank \#86

2026\-10\-10T23:01:00\.95027\+00:00: rank \#87

bigarrow gives agents a permission\-light way to point at UI elements, but the lone supplied reply questions whether human approval collapses into sudo\-clicking for the bot\.

## The brief

bigarrow is a small macOS CLI and agent skill that draws click\-through arrows, boxes and text over the screen to point a human at a UI element an agent cannot or should not click\. It targets elements by Accessibility labels, coordinates, windows or Peekaboo snapshots, raises the relevant app, and disappears on a timer, owner exit, hook or optional close button\. The project argues this fills a handoff gap for permission prompts, 2FA, CAPTCHAs and guided setup, while explicitly not clicking, typing or capturing\. It reports 87 automated tests and 17 behavior checks on clean macOS runners\.

- Drawing requires no Screen Recording or Accessibility permission; element and window targeting use Accessibility granted to the shell app, with exit code 4 signaling missing permission\.
- Commands include point, start/stop, elements and doctor; targets include \-\-at, \-\-rect, \-\-mouse, \-\-window App\[:title\], \-\-element Label \-\-app App, \-\-peekaboo ID and \-\-snapshot\.
- It claims click\-through except on the sign and shaft, never steals focus, works across displays, Spaces, full\-screen apps and Stage Manager, and hides when the target is covered\.
- Visual options include shapes bend/straight/zigzag/spiral, styles arrow/ring/box, sizes, colors, borders, \-\-follow, \-\-until\-click, \-\-say, \-\-png and \-\-dry\-run \-\-json\.
- Verification includes 87 automated tests plus 17 behavior checks; CI runs on macOS 15, with reported passes on macOS 26 and 27, and CPU measured at 1\.4% on a CI runner\.
- Chrome page elements work only with \-\-force\-renderer\-accessibility or VoiceOver; Electron apps work, and otherwise the tool points at page coordinates\.

## Discussion themes

Analyzed: 2026\-10\-09T23:00:47\.849106\+00:00

Analysis sample: Based on 15 of 18 usable stored comments. Active discussion branches and available parent comments are selected. The analysis input was further shortened to fit its context limit.

This sample may omit parts of the full thread. Selected themes do not measure community opinion or how common a view is.

### Overlay spoofing and agent permission guardrails

Commenters raise security concerns: an overlay that can draw on permission prompts could hide the decline button or alter approve copy; if a human only clicks sudo for an agent, giving root may be equivalent; Claude reportedly refuses passwords, security\-setting changes, and external account creation even in Yolo sudo, though custom auto\-mode classifier rules can be added or built\-ins disabled\.

Sources: [Comment 50019459](<https://news.ycombinator.com/item?id=50019459>) · [Comment 50018927](<https://news.ycombinator.com/item?id=50018927>) · [Comment 50018990](<https://news.ycombinator.com/item?id=50018990>) · [Comment 50019014](<https://news.ycombinator.com/item?id=50019014>)

### Reproduce agent\-built apps with your own agent

One reply proposes treating repos as markdown specs and having one's own agent implement them because trusting another agent's implementation is questionable; another says they point an agent at vibecoded apps to reproduce them with customizations\.

Sources: [Comment 50020182](<https://news.ycombinator.com/item?id=50020182>) · [Comment 50021430](<https://news.ycombinator.com/item?id=50021430>)

### Model drift threatens spec\-based reproduction

A reply warns that reproducing from specs may break over time due to model drift, requiring pinning to the intended model version or faithful emulation in a larger model\.

Sources: [Comment 50023755](<https://news.ycombinator.com/item?id=50023755>)

### Expensive, overengineered compute for solved tasks

Critics say the approach makes already\-solved tasks far more expensive, reliant on rented compute and massive electricity, and overengineered—like needing an external datacenter for videogame tutorials\.

Sources: [Comment 50022929](<https://news.ycombinator.com/item?id=50022929>) · [Comment 50024352](<https://news.ycombinator.com/item?id=50024352>)

### Useful for learning complex GUI applications

A reply argues the approach could still be useful for learning complicated GUI applications such as 3D modeling or media editing\.

Sources: [Comment 50024101](<https://news.ycombinator.com/item?id=50024101>)

### Feature gimmicks and 'slop' terminology

Comments joke about adding rainbow arrows, angry pointers, flame text boxes, and airhorn sounds, while a reply says the repo should be useful rather than give agents a blink\-tag\-like power; a separate tangent debates whether 'slop' is a verb\.

Sources: [Comment 50019128](<https://news.ycombinator.com/item?id=50019128>) · [Comment 50019211](<https://news.ycombinator.com/item?id=50019211>) · [Comment 50019339](<https://news.ycombinator.com/item?id=50019339>) · [Comment 50020047](<https://news.ycombinator.com/item?id=50020047>) · [Comment 50020705](<https://news.ycombinator.com/item?id=50020705>)

## Sources & coverage

AI-generated summary · 2026\-10\-09T13:01:54\.727012\+00:00

Based on 1 of 1 usable stored comments, selected by depth and branch activity. This is a sample of the discussion. Article text may also be shortened.

Generated using deepseek\-ai/DeepSeek\-V4\.1\-Flash. Check the linked sources for full context.
