locking-down-claude

name	locking-down-claude
description	Secure non-interactive Claude/SDK calls: pinned tools, prompts, permissions.

locking-down-claude

Rule: every programmatic Claude callsite sets four flags. Skip any one and a future edit silently widens the surface.

First: prefer the lib helper — don't hand-roll the flags

🚨 For Node scripts / hooks, use spawnAiAgent from @socketsecurity/lib-stable/ai/spawn with a tier from the AI_PROFILE ladder in @socketsecurity/lib-stable/ai/profiles. It enforces the four flags at the type level (SpawnAiAgentOptions requires tools / disallow / permissionMode), translates them per-agent (claude / codex / gemini / opencode), and owns --no-session-persistence, --add-dir, and the 529-overload retry. Hand-rolling a spawn('claude', [...flags]) is how the flag set drifts — and the prefer-async-spawn lint rule flags the raw spawn anyway.

import { AI_PROFILE } from '@socketsecurity/lib-stable/ai/profiles'
import { spawnAiAgent } from '@socketsecurity/lib-stable/ai/spawn'

const { exitCode, stdout } = await spawnAiAgent({
  ...AI_PROFILE.read, // or .edit / .create / .full
  : ,
  : repoRoot,
  :  *  * ,
})