Paste this into .claude/settings.json on day one of any Claude Code project:
"permissions": {
"deny": [
"Read(./.env)",
"Read(./.env.*)",
"Read(./secrets/**)",
"Edit(./.claude/**)",
"Write(./.claude/**)",
"Bash(git push *)",
"Bash(rm -rf *)"
]
}
Tip for agent-written PRs: comment "@coderabbitai verify".
CodeRabbit's new Verify beta runs your app, tests what the PR changed and posts results with recordings. Expected behavior comes from your docs and tests, not the new code.
Free during the beta.
https://t.co/GxfKJ4vBao
New from @Replit: Missions. Set an outcome, a spending cap and which actions need your OK. An agent swarm plans, builds and tests toward it while an AI judge checks the result.
Replit says it built its own monitoring platform in 2 days on a $3,000 cap.
What would you hand it?