Skip to content

cli.md

Latest commit

 

History

History
72 lines (58 loc) · 3.62 KB

cli.md

File metadata and controls

72 lines (58 loc) · 3.62 KB
Raw

The skill-tree CLI

One entry point for everything in here, from an ordinary shell (no Claude session needed). It exists for two reasons: the skills otherwise only document themselves inside Claude, and the commands behind them are scattered across scripts/ and skills/*/scripts/.

A bare skill-tree prints the whole surface — every command and every skill — because keeping track of it all is the point:

skill-tree                    # commands + skills (the default; `help` does the same)
skill-tree list               # just the skills, one line each
skill-tree list --json        # the same, machine-readable
skill-tree show <skill-name>  # print a skill's full playbook (--raw keeps frontmatter)
skill-tree doctor             # this checkout, dev-link state, which CLIs are wired up
skill-tree install            # re-run scripts/install.sh
skill-tree dev --on           # dev mode (see below)
skill-tree check              # validate every skill's frontmatter and bundled scripts
skill-tree stats              # how often each skill is invoked (--days N, --json)
skill-tree test               # the test suite

stats: is a skill earning its place?

  SKILL            USER  AGENT  LAST
  handoff            32     17  2026-09-29
  repo-audit         13      1  2026-10-05
  verify              0      0  never
  to-spec             0      -  never

Counted from Claude Code's own session transcripts (~/.claude/projects, or $CLAUDE_CONFIG_DIR/projects), so nothing has to be recording beforehand. USER is you typing /skill-tree:<name>; AGENT is the model loading it on its own. An AGENT of - means the skill sets disable-model-invocation, so 0 there is a skill the model could have reached for and never did.

Below the table it lists Skill calls that errored — a skill the model asked for and could not load. "Not in this repo" means some CLAUDE.md still names a skill that was renamed or deleted; the projects it happened in are listed so the reference can be found.

Two limits. Claude Code prunes transcripts after about 30 days, so "never" means "not in the days still on disk", and a skill added last week has had a week. And Copilot CLI keeps no comparable record, so only Claude sessions are counted.

Skills that ship their own CLI are reachable by name, with arguments passed straight through:

skill-tree screenshot latest
skill-tree handoff backlog    # this repo's backlog; --pick to choose another repo

handoff also gets a bare name on PATH (~/.local/bin/handoff) -- capturing a backlog item is a mid-thought action, and the longer form is enough friction to lose the thought.

Not every skill has one — verify and ui-designer are pure playbook. Those tell you so and point at skill-tree show <name>.

It's a dispatcher, not a reimplementation — each sub-command delegates to the script that already does the job, and propagates its exit code. install.sh puts it on PATH at ~/.local/bin/skill-tree.

Bash tab completion comes with it: install.sh links scripts/completions/skill-tree.bash into ~/.local/share/bash-completion/completions/skill-tree, which bash-completion loads on its own -- no shell rc is edited. It completes commands, skill CLIs, show <skill>, and each sub-CLI's own sub-commands and flags -- those are scraped from that CLI's --help at completion time, so they can't drift; where there's nothing to offer it falls back to filename completion. The candidates come from skill-tree __complete rather than a hard-coded list, so a new skill is completable the moment it exists. If your shell doesn't pick it up, source that file from your ~/.bashrc.