Is Cline safe?

Updated

Cline executes file edits + shell commands without per-step approval. Larger blast radius than autocomplete-only tools. Safe with explicit allow-list + scoped credentials.

TL;DR

Cline is an autonomous-agent VS Code extension. It executes operations without per-step user approval. The blast radius is correspondingly larger than autocomplete-only tools. Safe IF you (1) configure auto-approve to exclude destructive operations, (2) use dev-only credentials, (3) review Cline's commits via Securie before merge.

How it fails in production

Agent executes destructive operation

Same class as SaaStr-Lemkin-Replit + PocketOS-Cursor: AI agent runs DROP / TRUNCATE / rm -rf because it interpreted ambiguous instructions.

Agent installs malicious MCP server

Cline supports MCP. Apr 2026 Anthropic MCP RCE + tool-poisoning class apply.

.cline/ credential capture

Same dot-directory pattern as `.claude/`, `.cursor/` — captures local config including occasional credentials.

How to ship safely on Cline

  • Configure Cline auto-approve to exclude destructive operations (drop, truncate, delete, rm -rf)
  • Use separate dev credentials, never prod
  • Add `.cline/` to .gitignore + .npmignore
  • Set per-session spend caps
  • Securie reviews every Cline-committed change
What Securie covers

agent-scope crate enforces compile-time destructive-scope guards; mcp-guard validates Cline's MCP server interactions; secret_scanner catches `.cline/` artifact inclusion.

Verdict

Cline is safe with allow-list discipline + scoped credentials. The autonomous-edit blast radius is real — treat Cline like a junior engineer with admin access: scope it down, review every change.