Not Claude Code specific — @k9 decorator works with any Python agent (LangChain, AutoGen, CrewAI). For Claude Code it hooks via .claude/settings.json, zero code changes. For pi-agent: if it's Python-based, yes it works. https://github.com/liuhaotian2024-prog/K9Audit
GStack is a brilliant setup for maximizing Claude Code's velocity. But if you are letting an agent run autonomously across your repos, velocity without constraints is terrifying.
We recently had Case #001: a Claude Code agent got stuck in a 70-minute loop, repeatedly injecting a staging URL into a production config file. Raw logs showed "exit code 0" (all green).
To fix this, I built K9 Audit — a deterministic, non-LLM causal auditing layer. It drops directly into .claude/settings.json (zero code changes, perfectly compatible with GStack). It records a cryptographically hashed 5-tuple of what the agent did vs what it was supposed to do.
Not Claude Code specific — works with any Python agent via a one-line decorator. LangChain, AutoGen, CrewAI, or anything custom. Claude Code just gets a zero-config hook via .claude/settings.json. Pi-agent should work fine if it's Python-based.
Fair point — it's Python-based if you use the @k9 decorator directly. For Claude Code specifically, the hook works regardless of what language the agent is written in, because it intercepts at the tool call level via .claude/settings.json.
The role decomposition — distinct specialist personas for planning, review, QA, shipping — is the pattern worth paying attention to here, separate from the LOC debate. Then making them work together in a panel/team format.
I've been working on agent workflow tooling and what keeps surprising me is how quickly authoring the skills becomes the trivial part compared to keeping them composable and in sync across projects.
gstack's git-clone-and-copy install model works great solo, but I'm curious how it holds up when different repos need divergent review gates or QA flows — that's where every "just copy the skills folder" approach I've seen starts to buckle.
13 comments
[ 2.8 ms ] story [ 39.3 ms ] threadYou essentially get an agent in conductor.build who drafts multiple choice replies to your product and engineering questions from claude.
Dramatically improved code quality and speed of development for me.
and thank fuck for that
We recently had Case #001: a Claude Code agent got stuck in a 70-minute loop, repeatedly injecting a staging URL into a production config file. Raw logs showed "exit code 0" (all green).
To fix this, I built K9 Audit — a deterministic, non-LLM causal auditing layer. It drops directly into .claude/settings.json (zero code changes, perfectly compatible with GStack). It records a cryptographically hashed 5-tuple of what the agent did vs what it was supposed to do.
If you're using GStack to speed up, use K9 Audit as your seatbelt. Repo: https://github.com/liuhaotian2024-prog/K9Audit
https://registry.gitagent.sh/agent/shreyas-lyzr/gstack-agent
.
I've been working on agent workflow tooling and what keeps surprising me is how quickly authoring the skills becomes the trivial part compared to keeping them composable and in sync across projects.
gstack's git-clone-and-copy install model works great solo, but I'm curious how it holds up when different repos need divergent review gates or QA flows — that's where every "just copy the skills folder" approach I've seen starts to buckle.
Is this not a backdoor way for YC to get signal on what people are building and find more startup ideas?