AGENT.md

The package manager for AI coding-agent rules.
Latest Updates Here:

[AGENT.md @agent_dot_md]

Your AI coding agent obeys one file. Nobody reviews it.

Dead paths. Leaked keys. Hidden instructions.
In AGENTS.md, CLAUDE.md and skills, read on every prompt.

Find them in 10 seconds:
npx activate-agentmd audit your/repo

Free and open source.

#AI #Coding #OpenSource

[AGENT.md @agent_dot_md]

Code (MIT): https://t.co/mGqDttANcx
Site: https://t.co/sRCHsqbCkW

[AGENT.md @agent_dot_md]

Your AI coding agent obeys one file. Nobody reviews it.

Dead paths. Leaked keys. Hidden instructions.
In AGENTS.md, CLAUDE.md and skills, read on every prompt.

Find them in 10 seconds:
npx activate-agentmd audit your/repo

Free and open source.

#AI #Coding #OpenSource

[AGENT.md @agent_dot_md]

@iocryposto1273 Free CLI: npx activate-agentmd audit owner/repo
Team approvals: https://t.co/MMjGAhJEr1
Code: https://t.co/mGqDttANcx

[AGENT.md @agent_dot_md]

Your coding agent follows AGENTS.md. Who approved what it says?

Agent.md audits the file,
ties every change to a reviewer's approval,
and signs a record anyone can verify offline.

Audit your repo now👇
#AI #Coding #OpenSource

[AGENT.md @agent_dot_md]

@iocryposto1273 @falkordb @WeMakeDevs The scariest AI agent isn’t the one that breaks rules. It’s the one that follows every rule and still leaks your secrets.

[Aaditya Rawat @iocryposto1273]

My AI agent leaked a secret. Every guard said "fine."

Session 1 read .env. Session 2 sent it out. Both looked clean.

VIGIL remembers where every value came from in @falkordb and stops the leak before it runs.

6/6 caught. 0/6 without the graph.

@WeMakeDevs #AIAgents

[AGENT.md @agent_dot_md]

npx activate-agentmd install claude/Motion/lumen

Then ask Claude Code for a launch film. The full film is pinned on my profile.

https://t.co/mGqDttANcx

[AGENT.md @agent_dot_md]

This launch film was made by Claude Opus 5.5.

No editor. No After Effects. One markdown file of motion rules: easing curves, cuts on the beat, one idea per screen.

The film is HTML rendered frame by frame.

The standard is open source 👇

#AI #Coding #OpenSource

[AGENT.md @agent_dot_md]

@TH33ORACL3 That's the point. Confidence doesn't make outdated context correct. Read the repo first, then generate the code. 🎯

[AGENT.md @agent_dot_md]

Your coding agent still writes middleware.ts.

Next.js 16 renamed it to proxy.ts. The old file isn't picked up.

params is async now too.

The model's training data is older than your package.json.

So I made the CLI read your package.json 👇

#AI #Coding #OpenSource

[AGENT.md @agent_dot_md]

@Shipit_okx That’s the frustrating part: the code can look perfectly reasonable until you try to run it. Version-aware guidance can catch that mismatch much earlier.

[AGENT.md @agent_dot_md]

@sing123890 Well put. The goal isn't just to generate code that looks correct, but code that's correct for the environment it will actually run in.

[AGENT.md @agent_dot_md]

@spaceone_labs Exactly. A smarter model helps, but the right context helps it make decisions that actually fit the project it's working in.

[AGENT.md @agent_dot_md]

@FlowSenital 100%. Generating code is cheap. Figuring out why plausible code doesn't work with your actual stack is where the time disappears.

[AGENT.md @agent_dot_md]

@iocryposto1273 Exactly. In a monorepo, one stale assumption can spread across multiple packages. The agent needs to understand not just the codebase, but the versions each part actually runs on.

[AGENT.md @agent_dot_md]

@TH33ORACL3 Exactly. Tools should adapt to the project’s actual dependencies instead of relying on the model’s outdated training data. Version-aware coding should be the default, not an upgrade.

[AGENT.md @agent_dot_md]

npx activate-agentmd pins

Prints the breaking changes for the exact majors you run (Next 16, React 19, Prisma 7, Zod 4…) so your agent follows them over habit.

https://t.co/mGqDttANcx

[AGENT.md @agent_dot_md]

@nark3d That's the key distinction: paths can be regenerated, but intent can't. CI can catch a dead file reference; it can't tell you whether “never touch billing” is still the right rule.

The hard part is keeping both accurate. What checks do you run in CI?

[AGENT.md @agent_dot_md]

Your CLAUDE.md says API handlers live in src/api/handlers.ts.

That file doesn't exist anymore.

Your agent doesn't know. It believes the file, then guesses.

A stale instruction is worse than no instruction.

#AI #Coding #OpenSource

How old is the oldest line in yours?

[AGENT.md @agent_dot_md]

@cascadeaiscout Agreed. But letting an agent rewrite its own instructions unchecked can make a bad assumption permanent.

I'd go with: agent proposes, human approves, CI validates.

Do you gate these updates today, or let them land automatically?

[AGENT.md @agent_dot_md]

@deeponailabs Exactly. Paths and function names are the easiest to catch. I'm curious what you check next outdated commands, architectural assumptions, or something else?