safe-chains 0.210.2

Auto-allow safe bash commands in agentic coding tools
Documentation
[[command]]
name = "ssh-agent"
description = "Starts an SSH authentication agent in the background. Prints shell-eval output describing the socket and PID so the user can `eval $(ssh-agent)`. The agent is a long-running daemon. Carve-out: --help / --version / -k (kill) / -s/-c (output format) only."
url = "https://man.openbsd.org/ssh-agent"
researched_version = "OpenBSD-current 2026-05-27"
level = "SafeRead"
bare = true
eval_safe = true
eval_safe_flags = ["-c", "-k", "-s"]
examples_safe = [
    "ssh-agent",
    "ssh-agent -s",
    "eval \"$(ssh-agent)\"",
    "eval \"$(ssh-agent -s)\"",
    "eval \"$(ssh-agent -s -k)\"",
]
examples_denied = [
    "eval \"$(ssh-agent -D)\"",
    "eval \"$(ssh-agent -d)\"",
    "eval \"$(ssh-agent --help)\"",
    "eval \"$(ssh-agent -a /tmp/sock)\"",
]
standalone = [
    "--help", "-?", "-D", "-c",
    "-d", "-h", "-k", "-q", "-s",
    "-v",
]
valued = ["-E", "-O", "-P", "-a", "-t"]