Insight

[Insight @PirateInsights]

Free proxy lists are ~all dead. And some of the "working" ones inject scripts, break TLS, or are honeypots.

I built an open-source tool that checks them for real.

proxy-scraper (MIT, Python):

→ pulls from 700+ public lists
→ real HTTP / SOCKS4 / SOCKS5 handshake for every proxy
→ verified TLS test
→ honeypot + script-injection detection
→ hourly-updated list, per-country lists
→ local rotating proxy: one endpoint, failed proxies get swapped automatically
→ MCP server, so AI agents can fetch pages through checked proxies

Try it, no install:

uvx proxy-scraper-cli --pick 5

Honest limits: they're still free public proxies. Good for testing and public data, not for logins or anything private.

Built with AI assistance (Claude Code); I design and test it myself.

Feedback welcome: which check would you add?

#Python #WebScraping #opensource

[Insight @PirateInsights]

My free proxy checker dropped about 800 proxies an hour that had just answered a request. Turns out a third were only flaky: a second try 20 s later took the hourly list from 768 to 1,177.

MIT, runs hourly on GitHub Actions:
https://t.co/OQEqu9HxMD

[Insight @PirateInsights]

It's lower than it sounds: 69% means at least one AI-signed commit in 90 days, and a single one counts. The spread:
▎ → 46% have 5+ signed commits
▎ → 34% have ≥10% of commits signed
▎ → 19% ≥25%, 10% ≥50%
▎ → median repo: 3.5%
▎
▎ So agents touch most big projects, but in most of them they still write a small (disclosed) share. And that's a floor, since trailers get stripped.

[Insight @PirateInsights]

How much of popular open source is written with AI agents?

I measured disclosed agent signatures (co-author trailers, bot authors, PR footers) in the 1,000 most-starred active GitHub repos:

→ 69% have at least one AI-signed commit
→ 16% of all commits carry a signature
→ Claude Code shows up in 618 repos, Copilot in 185, Cursor in 182

But signatures get stripped. My own proxy-scraper shows 13% signed. A new experimental estimate (commit style vs. the repo's own history, with an interval) says 40–73%.

Open source: web, CLI, GitHub Action, README badge.

https://t.co/IDOJbM6vae

[Insight @PirateInsights]

@animus_aeternal Gracias. Sin la barra de error, el número sería marketing; con ella, es una medición

[Insight @PirateInsights]

Agreed on the error bar. One correction so it doesn't spread: 40–73% is the estimate for my own repo, not the top 1,000 overall. The estimate only runs where a repo already discloses agent use (≥10 signed commits), and even then about half the repos come back "too uncertain". So for most projects the honest answer is "at least X%, upper bound unknown". For enterprises it's the same lesson: trailers and telemetry give a floor, never the level.

[Insight @PirateInsights]

Exactly, good detail. With COMMIT_MESSAGES (the default) the trailers survive, which matches the data: squash commits are signed more often. With PR_BODY the commit gets the PR description, so a "Generated with Claude Code" footer survives but the commits' co-author lines don't. With BLANK everything is gone. That's why the new fix doesn't rely on the commit at all: if the squash commit "(#123)" has no signature, howmuchai looks up merged PR #123 (matched via merge_commit_sha) and counts it when the PR is signed. That covers PR_BODY and BLANK repos, at least when the signature is in the PR.

[Insight @PirateInsights]

Yeah, that's why 16% is explicitly a lower bound: it only counts what's disclosed. For repos that strip it there's an experimental range; my own repo is 13% signed, likely 40–73%. Also just shipped an MCP server, so you can ask Claude or Cursor "how much of this repo is AI?" directly. Curious: do you strip it for cleaner history or because of the stigma?

[Insight @PirateInsights]

You were right to ask, so I checked: squash commits are actually signed more often (17.5% vs 14.6%), since GitHub keeps co-author trailers when squashing. But 1,712 squash commits in 203 repos lost the trailer while their PR kept the "Generated with Claude Code" footer. Just shipped a fix: a commit "(#123)" now counts when merged PR #123 is signed, matched via merge_commit_sha (10/10 spot checks correct). Moves the overall number from 16.2% to ~17%. Thanks for the idea 🙏

[Insight @PirateInsights]

@acadictive Lets Connect

[Insight @PirateInsights]

Time to promote your product

share your startup url 🔥
#buildinpublic #tech #x

[Insight @PirateInsights]

Update: v0.2.1 fixes the web app on GitHub Pages (it didn't load before, sorry if you tried). New: a demo mode, so you can see the whole flow without a phone: https://t.co/RnXaBYSc71

[Insight @PirateInsights]

Android phones now download AI models you never asked for. AICore (Gemini Nano) alone can take up to 7 GB.

RemoveAndroidAI: plug in your phone, open a web page, click once. Gemini, AICore and Galaxy AI are off. No root, every step undoable. Open source.

https://t.co/krgaToiU5J

[Insight @PirateInsights]

Try it in Chrome or Edge (WebUSB, nothing installed or uploaded): https://t.co/muPqVVUlqD

Covers Pixel, Samsung, Xiaomi, Motorola and OnePlus. Every package links its sources and says what else stops working. v0.1, real-phone reports welcome.

[Insight @PirateInsights]

hexvero 0.12 adds `audit`: checks one Electron build for shell.openExternal with unchecked URLs, a missing or grant-all permission handler, and a missing or weak CSP. Each with file:line, SARIF output and CI exit codes. Never runs the app.

npx hexvero audit path/to/app.asar
https://t.co/zWP1aAMFn2

[Insight @PirateInsights]

Real example: Electron Fiddle 0.38 → 0.39, read from the two release ZIPs. Electron 40.1.0 → 40.8.5, no fuse moved, an ASAR integrity record that ships but whose validation fuse is off, plus a new IPC payload field, setting and endpoint.

https://t.co/fzc0Ym3iVr

[Insight @PirateInsights]

What did your last Electron release change for the pages inside it?

hexvero diffs two builds without running either: new IPC channels, preload members, window settings, fuses, network hosts. File and line on both sides, plus a CI gate.

npx hexvero demo
https://t.co/wJIOAO5gy4

[Insight @PirateInsights]

Anyone next week on the It-sa in Nuernberg?

[Insight @PirateInsights]

@Akhnatouch I had it already on tuesday. Incredible... Is the 20x really 20x?

[Insight @PirateInsights]

Guys I found a helpful tool for your work with docker:
Find out where Docker's disk space actually went, and what prune would NOT give back. Read-only, one command.

https://t.co/Ihorq97X06

[Insight @PirateInsights]

only claude code is the one you need

[bABA @BabaDev_]

If you could only keep one for the next year, which one?