i turned my subtitle tool into an agent skill

2026-07-03 14:04

i turned my subtitle tool into an agent skill built a subtitle recognition saas last year. openclaw dropped and it was dead before launch. a lot of people asked why i let it die. general agents just ate the use case. simple workflows got replaced overnight, and the tools behind them flipped from user-facing products to agent. no ui, no onboarding, just something an agent can read. subtitles are the clearest example of this. when a person does the work you need a real product. when an agent does it you need a skill it can follow. so i built one. give it audio + the original script. offline whisper handles the transcription, an llm figures out the timestamps and proofreads against the original, and it gives you back an srt file. 🔗 github.com/haoyiyin/easy-subtitle

Google made Gemini 3.5 Flash free.

No credit card, no nothing. Head to aistudio.google.com, create an API key, and plug it into any client with Gemini provider support. It's their strongest model right now. 1M token context. Text, images, audio, video all go in. Free tier rate limits are dynamic but fine for daily dev use. The catch: anything you send through the free tier gets used for training, so don't feed it anything sensitive.

2026-07-04

There's a Chinese Codex alternative getting some heat lately.

Tencent's WorkBuddy. Non-technical roles are buzzing about it. I tried it: basically Codex minus the install headache. No API keys, no subscription setup. Just open it and it works. Agent backend is reportedly OpenClaw in Sandbox. Free tier gives you 500 credits a month while they're running the trial giveaway. Desktop app. You talk to it in natural language. Merge spreadsheets, build a slide deck, clean up folders, write reports, search competitors. Comes with a library of user-shared agents and skills. Scheduled automation works. You can connect it to outside messaging tools too.

2026-07-03

Claude appeal rejected.

Won't refund the subscription either. I'm half convinced the appeal email is itself phishing, because I'm pretty sure they log the IP. DHH came to Shenzhen recently and showed how he uses Claude Code over there. That's the safest way now, honestly: run Claude Code on an overseas machine, connect through Tailscale, never use it locally. So I need to rent an overseas box with a residential IP, and register and log into Claude there.

2026-11-14

Sharing my current mobile only workflow.

Until late September it was the usual setup: parked in front of my laptop, terminal open, bossing pi around, fanning out subagents, then eyeballing diffs myself and filing the PR. Glued to the chair all day, every day. Late September I moved the whole thing to my phone, Hermes runs the show, pi takes the orders. My back is very grateful. Three weeks in, still no regrets. Stack: Hermes on flagship brains (gpt-6-astra / opus-5.5), pi on mid-tier (ds-v4.1-flash / sonnet / terra) with as few plugins as possible, the two talking over a2a. Hermes reads the issues, slices them into small tasks, spins up a worktree for each one, and dispatches a pi into it. When a pi finishes it pings Hermes, who decides: ship the PR or bounce it back for rework. I get back from a full day of being a tourist, or just wake up, and a dozen PRs are sitting there done. Pretty sweet. The catch is how black-box the whole thing is. My emotional arc has been a rerun of early vibe coding, except back then I at least read the diffs. Now even that goes to a flagship model. Boxes all the way down. If you don't have data anxiety, grok-bot and chatgpt-dot basically ship this out of the box, grok-cli and codex built in, native agent-to-agent comms. Downside: your code lives on their servers, everything's moderated, and one ban takes it all with it.

2026-11-14