nasuy

Building Chro: co-knowledge base for human and AI, multi AI agents locally from your ideas, notes, and research. https://x.com/n_asuy

Japan Joined on Jan 1, 2026 https://chro-ai.com

nasuy

Mar 26

18,883 MCP servers. Five Chinese tech giants joined this week. Zero security audits.

#ai #security #mcp #supplychainsecurity

3 min read

Want to connect with nasuy?

Create an account to connect with nasuy. You can also sign in below to proceed if you already have an account.

Create Account

Already have an account? Sign in

nasuy

Mar 24

"The human might be asleep." One line in Karpathy's program.md started 100 automatic experiments per night.

#ai #opensource #machinelearning #agents

4 min read

nasuy

Mar 20

The most valuable AI agent skills are buried in GitHub repos

#machinelearning #ai #agents #opensource

4 min read

nasuy

Mar 19

[Meta-RL] We told an AI agent 'you can fail 3 times.' Accuracy went up 19%.

#ai #agents #reinforcementlearning #machinelearning

4 min read

nasuy

Mar 18

[EvoSkill] An AI agent learned from its own failures and got 12 points more accurate.

#machinelearning #aiagentsclaudecode

4 min read

nasuy

Mar 17

Codified Context: A chemist wrote 100K lines of game code alone. The secret was the context architecture.

#agents #ai #architecture #productivity

3 min read

nasuy

Mar 15

MCP tool spoofing succeeds 100% of the time. A new paper maps 12 security risks across 4 agent protocols.

#agents #ai #mcp #security

3 min read

nasuy

Mar 9

AI Agents Don't Need to Touch the UI. WebMCP Is the Third Way.

#webmcp #mcp #aiagents #webdev

2 min read

nasuy

Mar 7

The secret isn't the model. It's the harness.

2 min read

DEV Community

nasuy

Badges

1 Week Community Wellness Streak

Writing Debut

18,883 MCP servers. Five Chinese tech giants joined this week. Zero security audits.

Want to connect with nasuy?

"The human might be asleep." One line in Karpathy's program.md started 100 automatic experiments per night.

The most valuable AI agent skills are buried in GitHub repos

[Meta-RL] We told an AI agent 'you can fail 3 times.' Accuracy went up 19%.

[EvoSkill] An AI agent learned from its own failures and got 12 points more accurate.

Codified Context: A chemist wrote 100K lines of game code alone. The secret was the context architecture.

MCP tool spoofing succeeds 100% of the time. A new paper maps 12 security risks across 4 agent protocols.

AI Agents Don't Need to Touch the UI. WebMCP Is the Third Way.

The secret isn't the model. It's the harness.