Thorim makes Claude Code and Cursor push back on dangerous asks
At a glance
An Indie Hackers founder almost shipped a no-auth “delete all users” endpoint Claude cheerfully wrote for a supposed test env—then built Thorim, a one-command layer of specialist agents/skills that makes coding agents challenge unsafe requests. This workflow is listed under Claude-Code, Cursor, Security, Devtools, Guardrails.
CircleCI's Chunk sidecars run scoped microbuilds in Firecracker/E2B microVMs that mirror CI, hooking agent stop/eval events so Claude Code, Codex, or Cursor changes are validated before commit—~27s microbuilds and 3–5× lower token burn on retries in their tests.
Room Service 3.3’s local MCP gives coding agents disk/port/Git visibility and actions like stop-port or trash build data—off by default, local socket only, permanent actions need in-app approval, every action logged.
A developer using both daily reports code quality converging once plans are clear, but keeps Cursor for slow focused diffs/sense-making and Claude Code for parallel/async agents plus well-scoped CI/CD tests and refactors.
Local-first Apache-2.0 CLI for Claude Code/Codex/Cursor: mcp-xray ranks bloated tool schemas, sync compresses CLAUDE.md↔AGENTS.md↔Cursor rules, and shield fails CI on leaked keys/prompt-injection/Trojan Unicode—claims large cuts in model-visible prompt size without API calls.