skillterm vs mods
LLM-piping shell tool (Go, Charmbracelet)
mods is a polished, Unix-flavoured way to put an LLM in a pipeline: `cat foo.json | mods "summarise this"`. skillterm is not pipe-shaped — it is a generator of structured per-tool knowledge and the completion hook that uses it. The right pick depends on whether you want a sharp pipe-friendly LLM client or a persistent skill layer.
mods rows describe publicly documented behaviour: stdin/stdout piping, Markdown output, multiple LLM backends, Glamour-rendered output. skillterm rows come from neul-labs/skillterm. We do not quote specific backend lists for mods because they evolve; rows that depend on backend choice are marked qualitatively.
| Dimension | skillterm | mods | Advantage |
|---|---|---|---|
| Primary surface | Generate SKILL.md, serve at Tab | Pipe stdin through an LLM, render Markdown to stdout | Comparable |
| Implementation language | Rust | Go | Comparable |
| Install | cargo install skillterm | Go install or Homebrew | Comparable |
| LLM access pattern | Headless agent runtime with web search, web fetch, bash, file I/O | Direct API to a chosen LLM backend | skillterm |
| Persistent artefact | SKILL.md per tool | Output is per-invocation | skillterm |
| Shell tab integration | init script for zsh / bash / fish | Designed for pipes, not completion | skillterm |
| Markdown rendering | Not the focus — SKILL.md is the artefact | First-class Markdown rendering of output | mods |
| Pipe-friendliness | Generation is interactive; not a pipe stage | Designed to live in shell pipelines | mods |
| Bootstrap skills concept | skill-creator and saas-detector | No equivalent | skillterm |
| Shareable result | Publish skills to a registry; install by name or gh:user/repo | Pipe stage; share scripts/aliases | skillterm |
| License | MIT | MIT | Comparable |
Pick skillterm when
- ▸You want per-tool SKILL.md as a persistent artefact, not a one-shot pipe
- ▸You want shell tab completion to surface tool-specific knowledge generated by an agent
- ▸You want the same artefact to be portable to a Claude-compatible agent runtime
- ▸You want a registry of shareable skills
Pick mods when
- ▸You want a Unix pipe-friendly LLM client: stdin in, formatted text out
- ▸You want polished Markdown rendering and beautiful terminal UI as a first-class concern
- ▸You want flexible backend choice and the option to point at multiple providers
- ▸You want a one-shot tool that fits into existing shell pipelines without a per-tool generation step
They often compose.
It is reasonable to use mods for one-shot prompts and skillterm for persistent per-tool completion in the same shell. They occupy different niches in the AI-CLI stack.