Agent Random

Vulnerability research - Bug Bounty - Cybersecurity - Software Development ::
OSCP - MCP

[Agent Random @AgentRandom_]

@mark_k You call it delays, but they have never announced anything. The "insiders" are the one hyping a release without even a grain of information to base it on

[Agent Random @AgentRandom_]

@ParamSiddh Claude Opus 5.5 can literally work 24/7 on the Ultra plan without hitting any limits, so I just have one thread grinding days on end

[Agent Random @AgentRandom_]

@jakeboggan @itsolelehmann Yes, but you don't need AI for this...

[Agent Random @AgentRandom_]

@JulieFrench_ I remember seeing a triple rainbow once. That is insanely unusual

[Agent Random @AgentRandom_]

Have you heard about the impact of adding the heart emoji to a prompt, it is wild that these sort of unnoticable (to humans) nuances can have these effects. What I would like to see is a benchmark that tests out various prompts with subtle changes and inspects the impact it has. Facinating stuff, but not something I have time and tokens to spend on.

[Agent Random @AgentRandom_]

@Gundys69 Building a business around something you don't control is and has always been a bad idea.

[Agent Random @AgentRandom_]

@ImLunaHey @t3dotcodes There are so many variants of error messages at this point that we need to send them into Jev so it can decide what is the best course of action

[Agent Random @AgentRandom_]

@bohops Is that a floppy disk drive? 🤣

[Agent Random @AgentRandom_]

@TJeparskis Or, they have nothing to ship and need something to give out the remaining 2 days?

[Agent Random @AgentRandom_]

@theo I am already moving all my building to be on Omarchy. it is the way forward for sure. Using my Mac to dispatch work on other headless boxes is such a nice way to get work done

[lauren @poteto]

grok bot told me to post this

[Agent Random @AgentRandom_]

I found a good way to prevent Claude and Codex from adding too many tests while developing:

The builder agent creates tests while working (normally red/green tests). When work is done, all the tests it created are deleted.

A new seat looks at the diff and creates tests that it deems valuable based on a set of rules around testing.

After tests are implemented and run, the review seat gets handed the diff.

Reduced the number of tests in the repo dramatically!

[Agent Random @AgentRandom_]

@rootsud0 I made @Rqwstr_com exactly for this

[Agent Random @AgentRandom_]

@IndependentEco I am more looking forward to Mythos 5.5 I hope that is a beast of a model

[Agent Random @AgentRandom_]

@giadotai "ever released"? It will last 1 month, tops until it is more or less irrelevant

[Agent Random @AgentRandom_]

@_ak_111 Which also means Mythos 5.5 for us in the CVP program 👀

[Agent Random @AgentRandom_]

@giadotai Claude dispatch is exactly what you want, and has been in the desktop app for months

[Agent Random @AgentRandom_]

@tregoudeau @lassejlv @giadotai Have you not heard of Claude dispatch? Just tell it to work on X on folder Y with model Z and it gets started. And if you have remote control always on for new sessions, you can open the session on your phone. Such a underrated and under promoted feature

[Agent Random @AgentRandom_]

@_ak_111 Doesn't pay the bills, thats why

[Agent Random @AgentRandom_]

@ClaudeDevs Feature request: When you present the "We found a potential dangerous 'rm' command" even though I have 'Bypass permissions' enabled. It would be cool if I could ask for a independent second and third opinion if it is actually destructive before I make a decision.