Shell completion¶

kubeagent completion <shell> prints a completion script, generated straight from the command tree — every subcommand, every flag — so it can never drift from what the binary actually accepts the way a hand-written script would.

kubeagent completion bash
kubeagent completion zsh
kubeagent completion fish
kubeagent completion powershell

It is the one kubeagent command that touches nothing at all: no cluster client, no kubeconfig, no model call. It works with KUBECONFIG unset, pointed at a nonexistent file, or with no network reachable — the script is built from the command tree in memory, not from anything kubeagent reads.

Installing it¶

Bash¶

Requires the bash-completion package. Load it for the current session:

source <(kubeagent completion bash)

Or install it for every new session:

kubeagent completion bash > /etc/bash_completion.d/kubeagent   # Linux
kubeagent completion bash > $(brew --prefix)/etc/bash_completion.d/kubeagent   # macOS

Zsh¶

If shell completion is not already enabled, turn it on once:

echo "autoload -U compinit; compinit" >> ~/.zshrc

Load it for the current session:

source <(kubeagent completion zsh)

Or install it for every new session:

kubeagent completion zsh > "${fpath[1]}/_kubeagent"   # Linux
kubeagent completion zsh > $(brew --prefix)/share/zsh/site-functions/_kubeagent   # macOS

Fish¶

Load it for the current session:

kubeagent completion fish | source

Or install it for every new session:

kubeagent completion fish > ~/.config/fish/completions/kubeagent.fish

PowerShell¶

Load it for the current session:

kubeagent completion powershell | Out-String | Invoke-Expression

To load it for every new session, add that same line to your PowerShell profile ($PROFILE).

Every shell needs a new session (or a re-sourced profile) before the change takes effect.

Under krew: a script alone is not enough¶

If kubeagent is installed as the kubectl plugin kubectl kubeagent (see Install), the sections above are necessary but not sufficient — read this before assuming source-ing the bash or zsh script is enough.

Cobra's completion generators — the code that produces the scripts above — read the command's plain name, never the display-name annotation that makes kubectl kubeagent the spelling kubeagent uses in its own usage and error text. That was verified directly: building the binary under both names and diffing the two generated scripts shows they are byte-identical, and both register completion for the bare word kubeagent. So sourcing the script gives you tab completion when you invoke the binary directly as kubeagent. It gives you nothing for kubectl kubeagent <TAB> — kubectl does not consult a plugin's own completion script at all.

kubectl instead looks for a separate executable named kubectl_complete-<plugin> on your PATH, and runs that to answer a completion request. This is a different mechanism from the script above, not another way of installing it — kubeagent does not generate this file for you; it is a one-line shim you create once, that forwards the request back into kubeagent's own __complete command:

cat > ~/.krew/bin/kubectl_complete-kubeagent <<'EOF'
#!/usr/bin/env sh
kubectl kubeagent __complete "$@"
EOF
chmod +x ~/.krew/bin/kubectl_complete-kubeagent

~/.krew/bin is already on PATH if you followed krew's own install instructions, so no further setup is needed once the shim is in place and executable. With it, kubectl kubeagent <TAB> completes; without it, krew users get no completion at all regardless of whether they also sourced the bash or zsh script above.