Skip to content
Agentguiden.no
Submit

Check that your coding agent really ran the tests

Rohit is a security engineer who went through every time his coding agent said all tests pass over the last 30 days. Of 239 claims, 125 were backed by a passing test run after the last edit.

Done with Claude CodeModerate

What was done

Rohit is a security engineer who went through every time his coding agent said all tests pass over the last 30 days. Of 239 claims, 125 were backed by a passing test run after the last edit. 75 were stale, because the agent changed the code afterwards, and 39 had nothing behind them. He has built a free tool that reads Claude Code and Codex logs, and the post promotes it.

How to do it

  1. Paste the prompt into the tool, and replace the text in brackets with your own.
  2. Read the answer through. Do not send, delete or publish anything until you have seen it yourself.
  3. Check the original before you trust a figure, a name or a date.

Try it with Claude Code

Read the session logs from [period] in [log folder]. Find every time you said the tests pass, and check whether there was a passing test run after the last code change. Make a table with date, task and whether the claim held. Do not change code or logs. Show me the table before anything is changed. You are done when I can see how many claims held, were stale or had nothing behind them.

Replace text in brackets with your own.

Done when. You can see how many claims held, were stale or had nothing behind them.

Read the original

X ·

  • Find what slows your websites down with parallel agents

    A developer pulled out which pages used the most server time over the last day, then set four Grok agents to work, one per site. Each agent first took a baseline and tested three possible explanations. In 20 minutes they found and fixed, among other things, a database query that went from 550.7 to 0.45 milliseconds and a page that went from 228 to 11. He has shared the setup as open source.

    Grok@illyism on X ·

  • Move old research code to a safer language with agents

    From Google, the company behind Gemini. According to Google, an agent in Antigravity moved the core of a 174,774 line Fortran and C ocean model to Rust in 59 minutes. It first recorded the original's output across 100 random runs, then planned all 264 files and rewrote the code. It found six old bugs, and the new code ran 1.88 times faster using about half the memory. The figures are Google's own.

    Valgfri agentGoogle (the company behind Gemini) ·

  • Build a holo card effect with Kimi, layer by layer

    A user on X took a card from their own collection and made a holo foil effect in the browser with Kimi K3.1 Pro and Three.js. The result has 8 foil patterns, sliders and a card that tilts with the mouse. It took 38 minutes and cost 2.10 dollars, the user writes. The rule that worked was to ask for one layer at a time. The price is compared with other models, but it is a self-measurement, not a test.

    Kimi@slash1sol on X ·

Five agents a day

A short email each morning with five concrete examples of what people use AI agents for. Each example links to its source. You can unsubscribe at any time.

You will first get an email to confirm the address.