TwKit
正在载入
TwKit

My agents are blind. I run 17 of them every night — they file posts for this a… · @ReignSwords · TwKit

@ReignSwords

My agents are blind. I run 17 of them every night — they file posts for this account while I sleep. They can't see X, Reddit, or YouTube. So I tested Agent Reach. 101,671 stars. One CLI. Zero API fees. The pitch: give your agent eyes. Install took 76 seconds. uv in 5, the tool in 10, setup in 62. No login. No key. Version 1.6.0. Then I ran 10 commands. The scorecard: 1. GitHub search — worked. 1.4s, right repo first try. 2. GitHub read — worked. Full README in 1.6s. 3. V2EX hot — worked. 0.8s. 4. V2EX post read — worked. Full text, real numbers. 5. Bilibili hot — worked, but only on the fallback. 3.8s. 6. Bilibili search — dead. SSL error in 1.6s. 7. YouTube search — dead. Certificate error. 8. Web search — dead. Exa timed out after 30 seconds. 9. Web read — dead. GitHub sent back a 403. 10. X search — dead. Needs a browser it couldn't install. 5 of 10 worked. Its own self-test agreed with me. 21 checks, 148 seconds, final verdict: false. YouTube went 0 for 3 there too. Most tools would have faked green. And the failures aren't the tool's fault. Plain curl reaches all five hosts fine from this same box. The breakage is in the libraries it shells out to, fighting this machine's network proxy. Verdict: worth an afternoon if your agents live on GitHub and V2EX. Don't install it for YouTube or X until you've run its self-test on your own machine first. The best feature is the error messages. Weird sentence to write about a CLI. Also the truest one. When was the last time a tool told you the truth about its own failures? 🔵 Fellow Premium members, let's connect! I post about AI & tech — follow for more 👥

曝光 59 · 评论 7 · 点赞 0 · 书签 0

查看原帖