Use --dangerously-skip-permissions for Fully Autonomous Claude in Trusted CI

Claude Code asks for permission before running commands, editing files, and using tools. In interactive sessions, that's a safety feature. In CI, it's a blocker — there's no human to click "approve". The --dangerously-skip-permissions flag removes the prompts entirely.

claude --dangerously-skip-permissions -p "run the tests and fix any failures"

Claude reads, writes, and executes without asking — fully autonomous. The flag name is deliberately alarming because this is powerful and should only be used in environments you trust.

A typical CI setup:

# GitHub Actions example
- name: Auto-fix failing tests
  run: |
    claude --dangerously-skip-permissions \
      --model haiku \
      --max-turns 20 \
      -p "run the tests, fix any failures, and commit the fixes"

Combine it with other safety controls:

# Limit tool access even without permission prompts
claude --dangerously-skip-permissions \
  --allowedTools "Read,Glob,Grep" \
  --max-turns 10 \
  -p "review this code for issues"

This gives you autonomous execution with guardrails — Claude can read and search freely but can't edit files or run arbitrary commands, even with permissions skipped.

Practical CI use cases:

# Automated code review on every PR
claude --dangerously-skip-permissions \
  --max-turns 15 \
  -p "review the changes in this PR for bugs and security issues" \
  --output-format json

# Auto-fix linting issues
claude --dangerously-skip-permissions \
  -p "fix all linting errors and commit the changes"

# Generate documentation on release
claude --dangerously-skip-permissions \
  -p "update the API docs to match the current endpoints"

Important safety notes:

• Never use this flag on untrusted input — it grants full system access
• Always combine with --max-turns to prevent runaway execution
• Use --allowedTools to restrict what Claude can do even in autonomous mode
• Only run in isolated environments (CI containers, sandboxed VMs)

Automation needs autonomy. --dangerously-skip-permissions gives it — but only use it where you'd trust a script with the same access.

via Claude Code