frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Show HN: Grepathy – Claude made a decision nobody approved

https://github.com/evansjp/grepathy
12•evansjp•1h ago
hey HN - Claude pre-created users in Clerk with null emails/names as "guest users" on a contract job. Wasn't in any plan. The CTO asked why, and I didn't know! I didn't make that decision!

The reasoning was in a transcript on my laptop. Claude Code deletes those after 30 days by default. Two of my projects lost their whole history that way.

Grepathy distills transcripts locally into markdown committed with the code. Decisions only, never your messages, no server.

List every decision nobody approved:

  grep -rn "agent-initiated" .ai/why/
Ran a blind eval before shipping, published it including the misses (REPORT.md). Agents with Grepathy answered the "why" questions right. Baseline agents made up confident wrong answers.

Comments

ballenf•54m ago
Good catch on CTOs part. That scenario is one that keeps me awake at night -- that some large PR of mine has come functionality or feature that I didn't ask for and didn't notice.

How did the CTO respond to "Claude did it and I didn't catch it"? Did the AI PR summary mention it? Or did the CTO just read the code?

evansjp•39m ago
CTO responded well to it - frankly because we're all using some coding agent and it's just the natural process of code review for them to sniff things out. No mention of it in any PR summary - CTO found it in the migration that would mutate the User record to accept `null` email.
hhh•28m ago
I am fighting against people doing this every day, and we've made it extremely clear that your computer and processes running on it are an extension of you, a human, who is responsible for the actions on your computer.
Synthetic7346•13m ago
Won't that slow you down a lot? I get it in principle, but in practice if you have to review everything your coding agent does the productivity gains diminish significantly
extrabajs•2m ago
I guess you get to choose between reviewing it yourself or wasting someone else’s time
sixtyj•53m ago
Well done.

This is not just Claude’s behaviour :)

It could be used for every agentic coding as well, because I have experienced such mess so many times in last three years.

One issue repaired, another one touched and changed…

evansjp•33m ago
thank you so much! started with a Claude Code adapter, but you think Codex would be a good next one?
q3k•51m ago
> I didn't know! I didn't make that decision!

Thanks for today's dose of impostor syndrome cure.

embedding-shape•47m ago
> Decisions only, never your messages

For me this is basically the opposite of how it would need to be, the decisions are all in my messages to the agent, almost never directly in the agent's replies, sometimes indirectly though I suppose.

Interesting to see that apparently some people use these tools and are the ones listening and doing what the LLM decide and says, rather than the opposite.

> Ran a blind eval before shipping, published it including the misses (REPORT.md). Agents with Grepathy answered the "why" questions right. Baseline agents made up confident wrong answers.

Is this eval public anywhere? The whole ecosystem severely lacks transparency, and people continue making strong claims without any sort of data or results that backs these up. Is there any public somewhere backing up the idea that it's more accurate that just using whatever agent harness?

> Claude made a decision nobody approved

It irks me a tiny bit that this is essentially clickbait, you don't know for sure Claude "made a decision nobody approved" as you've lost the data because you never thought sufficiently about storing these things in the first place. Most likely it misunderstood you at one point, you didn't carefully read the reply and approved it all, so most surely what you've done lead to that being done. These big models hardly ever just randomly do whatever, you can more often than not trace it to something you did wrong.

SpicyLemonZest•31m ago
Can we get the transcript of the session where you had Claude write this post? I want to understand the "why" behind this project, and whether a human being was at any point involved in the "blind, pre-registered evaluation of Grepathy against an honest baseline".
albert_e•23m ago
> The reasoning was in a transcript on my laptop. Claude Code deletes those after 30 days by default. Two of my projects lost their whole history that way.

Wait what? Claude Code (and Codex) transcripts are auto-deleted??

That's a treasure trove of information for all my projects and I had built some tooling and workflows tahat deal with extracting insights from them -- withut ever realizing I mightbe losing old chats. Thanks for this -- I may need to include an archival step for retention.

weird-eye-issue•19m ago
Yes but it's just a setting you can change I don't think you really need an archival step
trjordan•8m ago
> "cleanupPeriodDays": 99999

Throw that in ~/.claude/settings.json

vanyaland•22m ago
Your .ai/why/main.md is at 97KB and 148 entries. Does anything prune that, or does a long-lived branch keep growing?
trjordan•6m ago
100% important. But what decisions do you care about seeing?

The whole point of the agent is to make decisions for you. If you want to make every little detailed decision, just write the code.

The whole art of this problem is figuring out which decisions matter to you, and how to surface them.

(Disclosure: we're working on this too. https://tern.sh)

Show HN: I'm a Robot

https://castrio.me/im-a-robot
1•non-•55s ago•0 comments

GPU-Tile-SIM: Tile-Centric GPU Simulation for LLM Hardware-Software Co-Design

https://arxiv.org/abs/2607.11262
1•rbanffy•1m ago•0 comments

The Model Was Never the Hard Part

https://srirupa19.github.io/gsoc/2026/07/14/gsoc2.html
1•LorenDB•1m ago•0 comments

Getting Clients Cold Calling

https://www.fludileads.co.uk
1•olliehemps•1m ago•1 comments

How to explain LLM architecture to your mom and dad

https://www.ibm.com/think/news/what-does-ai-look-like
1•linerep43•2m ago•0 comments

Show HN: HeapLens – Java heap dump analysis inside VS Code

https://github.com/sachinkg12/heaplens
1•sachinkg12•3m ago•0 comments

Show HN: Stacks – an hourly card solitaire about building four gardens

https://plan9.kr/stacks/
1•sungchi•3m ago•0 comments

Agentmetry – An open-source flight recorder (SIEM) for AI Agents

https://github.com/blitzcrieg1/agentmetry
1•blitzcrieg1•3m ago•0 comments

Don't ask what you want. Ask who you want to be

https://softwaredoug.com/blog/2026/07/13/who-want-to-be.html
1•zdw•3m ago•0 comments

MemExchange: Utility-Driven Distributed Memory Reallocation for Datacenters

https://arxiv.org/abs/2607.11579
1•rbanffy•4m ago•0 comments

Industrial Revolution vs. AI Revolution: Same Fears, New Era

2•historical1234•4m ago•0 comments

MCP Queen – a graded registry that live-probes every MCP server

https://mcpqueen.com/
1•healthai•4m ago•0 comments

Publisher ad supply fell by up to 40% in Q2 as AI search choked the open web

https://digiday.com/media/publisher-ad-supply-fell-by-up-to-40-in-q2-as-ai-search-choked-the-open...
1•thm•5m ago•0 comments

I ran Sonnet 5 vs. Opus 4.8 head to head on 24 tasks to see what's different

https://www.stet.sh/blog/sonnet-5-vs-opus-4-8-reasoning-dial
1•bisonbear•5m ago•0 comments

Agent K

https://github.com/gabereiser/agent-k
1•reactordev•6m ago•1 comments

Design of automated systems must factor in human psychology – Knowable Magazine

https://knowablemagazine.org/content/article/mind/2026/design-automated-systems-with-human-psycho...
1•rbanffy•6m ago•0 comments

Show HN: I blind-test 6 LLMs daily by having them summarize the same story

https://www.snipvote.com
1•SnipVote•7m ago•0 comments

Only a Third of "Remote" Jobs Are Remote

https://blog.omnigrade.ai/blog/fake-remote/
1•mkysel•8m ago•0 comments

Choo Choo Has an Ouchie

https://jackcheng.com/sunday/459-choo-choo-has-an-ouchie/
1•herbertl•8m ago•0 comments

Ask HN: Have Chat AI trained you to use Shift+Enter?

1•markmatsushima•9m ago•0 comments

Microsoft's Secure Boot has been broken for a decade and no one noticed

https://arstechnica.com/security/2026/07/microsoft-secure-boot-has-been-broken-for-most-of-its-ex...
1•RattlesnakeJake•9m ago•0 comments

Show HN: Aproov – Reach for founders, deals for creators

https://aproov.es
1•Jaime_MB•9m ago•1 comments

OpenAI loses trademark dispute at EU court

https://dpa-international.com/economics/urn:newsml:dpa.com:20090101:260715-930-389143/
2•hermanzegerman•11m ago•0 comments

T. rex named Gus sells for $50.1M to become the most expensive dinosaur

https://www.nbcnews.com/news/us-news/t-rex-named-gus-sells-501-million-become-expensive-fossil-ev...
1•OptionOfT•11m ago•1 comments

Why "Yes" Is Overrated

https://www.votito.com/methods/acquiescence-effect/
2•gojkoa•12m ago•1 comments

In Tenuous Defense of War Markets

https://thebsdetector.substack.com/p/in-tenuous-defense-of-war-markets
1•surprisetalk•13m ago•0 comments

CVE-2026-59208: Cross-Issuer Account Takeover in n8n

https://www.strix.ai/blog/n8n-cross-issuer-account-takeover
5•bearsyankees•13m ago•0 comments

Show HN: Hillnote – local first Markdown workspace, AI via CLI, Ollama, MCP

https://hillnote.com
1•rajathbail•13m ago•0 comments

FreeBSD Laptop Support Continues Improving with WiFi, GPU and Audio Driver Work

https://www.phoronix.com/news/FreeBSD-Laptops-June-2026
2•Bender•14m ago•0 comments

Vocalinux 0.14 Beta Released for Offline Voice Dictation Speech-to-Text on Linux

https://www.phoronix.com/news/Vocalinux-0.14-Beta
1•Bender•15m ago•0 comments