← home · glossary
The whole vocabulary, defined.
skillterm's language is small: a skill format, two runtimes, two bootstrap skills, and a handful of commands. Learn these and every command reads plainly.
The skill format
- skill
- A directory containing a SKILL.md file, optionally with references/ and scripts/ siblings. The unit skillterm generates, installs, and serves at Tab time.
- see also: SKILL.md · skill directories
- SKILL.md
- The skill file: YAML frontmatter (at minimum name and description) plus a markdown body with overview, commands, examples, and best practices. Follows the Claude Agent Skills convention.
- see also: skill · skill-creator
Agent runtimes
- agent runtime
- The headless coding agent skillterm invokes to do the reasoning: Claude Code by default (runtime = "claude-code"), Codex as the alternative (runtime = "codex").
- see also: Claude Code · Codex · headless mode
- Claude Code
- The primary, recommended agent runtime. Invoked in headless mode; the agent uses WebSearch, WebFetch, Bash, and Read/Write to build a skill.
- see also: agent runtime · headless mode
- Codex
- The alternative agent runtime, configured with runtime = "codex". Same generation workflow and same SKILL.md output as Claude Code.
- see also: agent runtime
- headless mode
- Running the agent runtime non-interactively so skillterm can orchestrate it. The CLI is the orchestrator; the agent brings tool use.
- see also: agent runtime · tool use
Bootstrap skills
- skill-creator
- One of two bundled bootstrap meta-skills. Teaches the agent the SKILL.md format spec and a quality checklist. Improving skillterm largely means improving this skill.
- see also: saas-detector · SKILL.md
- saas-detector
- The second bootstrap meta-skill. Teaches the agent how to classify a tool and find its documentation before generating a skill.
- see also: skill-creator · web docs
Commands
- generate
- The core command. Builds a SKILL.md via the agent runtime from a system man page (default), a docs URL (--from-url), or open web search (--from-search).
- see also: agent runtime · complete
- complete
- Called by the shell hook on each Tab press. Returns contextual completions sourced from the matching SKILL.md. The cheap half of the tool: generation is expensive and one-time; completion is cheap and forever.
- see also: generate · shell hook
Core concepts
- shell hook
- The integration installed via skillterm init <shell>. zsh and bash hooks are eval-piped; the fish hook is sourced. It calls skillterm complete at Tab time.
- see also: complete · shells
- shells
- The three supported shells for completion integration: zsh, bash, and fish.
- see also: shell hook
- skill directories
- The three locations skillterm searches in order: project (./.skillterm/skills/), user (~/.skillterm/skills/), and system (/usr/share/skillterm/skills/). Project wins.
- see also: skill · registry
- registry
- The layer that makes skills shareable packages: search, install (by name, version, or gh:user/repo), and publish. Skills are versioned units of knowledge.
- see also: skill · publish
- tool use
- The built-in capabilities the agent runtime brings — WebSearch, WebFetch, Bash, Read/Write — that skillterm would otherwise have to re-implement around a raw LLM API.
- see also: agent runtime · generate