Quick Start
Launch your agent
aileron launch claude
Aileron spawns the agent as a child process with a policy-enforced shell. Every command the agent runs flows through aileron-sh and the policy engine before reaching the real shell. Aileron handles agent-specific quirks (shell validation, command wrapping) so the policy rules stay clean.
This works with zero configuration. Built-in defaults allow common toolchain commands (Go, Node, Python, Rust, Ruby, Elixir, Java) and deny dangerous operations (recursive delete, push to main). Every command flows through aileron-sh and the policy engine. Aileron is the single approval layer, replacing the agent’s native tool approval.
Policy evaluates every command
Safe commands (tests, builds, reads) auto-approve silently. Dangerous commands (force push, recursive delete) are hard-denied. Ambiguous commands (commit, push, deploy) prompt you once with context:
⏸ aileron: agent wants to run `git push origin feature/auth`
matched rule: ask (git push)
[y] allow [n] deny [a] allow always [s] show details
Scaffold a project policy
Optionally, create a project-specific policy:
aileron init
This creates a minimal aileron.yaml with project-specific deny rules and env scrubbing. Language toolchain and OS rules are built in, so you don’t need to list them.
See your config at a glance
aileron status # show everything
aileron status policy # merged policy: defaults + project + user
aileron status env # scrubbed and passthrough vars
aileron status notifications # Slack/Discord channels and token status
aileron status vault # stored secrets (names only)