Aakash

Building for 10+ years. Open source FTW! Metalhead.

Co-Founder - AATBit Labs

Connect with me at

[Aakash @aakashH242]

In case anyone's curious, dropping the link here.
https://t.co/Wfzhj5BCet

[Aakash @aakashH242]

Just released a new feature in Tenwrite that a few customers had requested. 😃

Tenwrite now suggests SEO titles, descriptions, focus keywords, categories, tags and image alt text for your publications.

Use them during manual export or via automations.

@TenwriteApp @rupampoddar_

[Aakash @aakashH242]

@abiodunosagie1 That’s a sneaky one. Computing the gain from elapsed sample count rather than subtracting a tiny amount each sample would avoid accumulated rounding error too.

[Aakash @aakashH242]

@OpenMinis For unattended runs, does a timeout trigger a fresh screen read before retrying a tap? A tap can succeed even when the response gets lost, so blind retries could submit the same action twice.

[Aakash @aakashH242]

@iamspiggle Where do you put portability in that list? Being able to export the whole app state and move later can matter even when the current hosting bill is tiny.

[Aakash @aakashH242]

@SamVale29 Even 'run tests' can execute arbitrary repo scripts. A disposable workspace with no production credentials gives you a much better safety boundary than an allowlist of command names.

[Aakash @aakashH242]

@hammer_mt Worth versioning the judge prompt alongside the task fixtures too. Then a score change can be traced to the agent, the grading rubric, or the input data without guessing.

[Aakash @aakashH242]

@CobaltHere Nice, that’s exactly the gap I was wondering about. Thanks for clarifying!

[Arsenal @Arsenal]

A spirited second-half comeback 💪

Watch the Full Match Replay from yesterday's 2-1 win over Leeds United now👇

[Aakash @aakashH242]

@VaishaliBuilds I’d compare against the full-context answer and raise the budget if key details go missing. I built Trimwise for query-aware trimming that keeps selected passages verbatim. Might be useful for your experiments: https://t.co/hTMpEq6J6E

[Aakash @aakashH242]

@danizhu The correction path would make a useful test: let two agents read a fact, correct it while one is still working, then check whether that agent can write the old version back. That seems easy to miss when testing memory one agent at a time.

[Aakash @aakashH242]

@RABSWFL Ah nice, thanks for clarifying! Keeping filenames and flags untouched was the bit I was checking.

[Aakash @aakashH242]

@KI_Vater Gauntlet’s an agent review workflow rather than a static scanner. It pushes the agent to back findings with evidence, group duplicates by root cause, and challenge suspected false positives. Source traces or runtime tests help separate real issues from noise.

[Aakash @aakashH242]

@fil_ships Looking forward to your feedback. Happy to improve it based on your use-case. 😃

[Aakash @aakashH242]

@fil_ships If it helps, you can check out a library I wrote specifically for this purpose - trimming context smartly instead of truncation.

https://t.co/hTMpEq7gWc

[Aakash @aakashH242]

@dezelin Ah got it, the branch shortcuts were the bit I was curious about. Thanks for explaining!

[Aakash @aakashH242]

@fil_ships What do you do if a ticket starts pushing past 150k? Split it further, or trim the context and keep going?

[Aakash @aakashH242]

@maheshboda006 Good to see a side-by-side test. The save step would be interesting to dig into: can annotations made in either build be reopened and edited in Preview? That'd say a lot about file compatibility.