← home · quickstart
Five steps take you from an empty shell to intelligent, documented completions on any command. You need Rust, cargo, and one agent runtime.
skillterm is a Rust binary. You also need one agent runtime on your PATH — Claude Code is recommended, Codex is the alternative.
# skillterm itselfcargo install skillterm # the recommended agent runtimenpm install -g @anthropic-ai/claude-code # ...or the alternativenpm install -g @openai/codex
Add the completion hook to your shell. zsh and bash are eval-piped; fish is sourced. Put the line in your shell rc file to make it permanent.
# zsh (in ~/.zshrc)eval "$(skillterm init zsh)" # bash (in ~/.bashrc)eval "$(skillterm init bash)" # fish (in ~/.config/fish/config.fish)skillterm init fish | source
Give the agent runtime its API key. skillterm reads per-runtime settings (binary path, model, timeout) from ~/.skillterm/config.toml.
export ANTHROPIC_API_KEY="sk-ant-..." # inspect / change configskillterm config listskillterm config test-provider
Ask skillterm to build a skill. The agent searches for docs, fetches them, runs man / --help, and writes a Claude-compatible SKILL.md.
# from a man page (default)skillterm generate kubectl # from a docs URLskillterm generate stripe --from-url https://docs.stripe.com # from open web searchskillterm generate terraform --from-search
Press Tab on the command. The shell hook calls skillterm complete, which serves contextual completions from the matching SKILL.md — cheap, every time.
kubectl get <TAB># → completions from ~/.skillterm/skills/kubectl/SKILL.md