Meta skills
Most skills teach your agent about your domain. Meta skills teach it about
Tome itself. They are bundled inside the tome binary — versioned with it,
with no download — and install as native SKILL.md files into your harness,
so the agent reads them the same way it reads any other skill.
List available meta skills
tome meta list
convert-marketplace — Guided conversion of a Claude Code marketplace into Tome's native plugin format. Drives `tome convert` and `tome lint` for the mechanical work, applies judgment to the parts Tome cannot represent, verifies the result, then reports to you and waits for explicit confirmation before registering anything in a workspace. [1a57b1e8df28caf7]
claude-code/global: not-installed
claude-code/project: not-installed
codex/global: not-installed
codex/project: not-installed
cursor/global: not-installed
cursor/project: not-installed
opencode/global: not-installed
opencode/project: not-installed
Each bundled skill appears with its revision (the hash in brackets) and an install state for every harness × scope pair Tome can target. The harness rows reflect the skill-capable harnesses on your machine, so your list may name more than the four above.
Gemini CLI has no native skill support, so it is not a meta-skill target — which is why it doesn't appear in the list above.
Install one
tome meta add convert-marketplace
By default this writes the skill into the project skill directory of every detected harness — run it once from your project root and every agent you use there can read it. Removal works the same way:
tome meta remove convert-marketplace
Choose the target
| Flag | Effect |
|---|---|
--harness <name> | Target specific harnesses instead of all detected ones. Repeatable: --harness claude-code --harness cursor. |
--global | Install into the harness's global skill directory instead of the project one. |
--force | Overwrite an existing install (including one that's been edited on disk). add only. |
--harness and --global work the same way on remove.
The convert-marketplace skill
This is the bundled skill to install first. It guides your agent through
converting a Claude Code marketplace into Tome's native format: it drives
tome catalog convert and tome catalog lint for the mechanical work,
applies judgment to the parts Tome cannot represent, verifies the result —
then reports to you and waits for explicit confirmation before registering
anything in a workspace. Your agent does the conversion; you give the final
approval.
The manual version of that workflow lives in Converting.
Keep installs up to date
An installed copy can drift from the binary. Tome compares the revision
stamped on the on-disk SKILL.md against the one bundled in the running
binary, and a copy whose revision no longer matches is stale — whether a
Tome upgrade shipped a newer revision, or someone edited the installed file, or
the file went malformed on disk. There is no separate "modified" state; an
edited copy is a revision mismatch like any other, so it reads as stale.
tome doctor reports every stale meta-skill install, and
tome doctor --fix
refreshes each one in place from the bundled copy. A location where the skill
was never installed is not-installed, not stale, so doctor leaves it alone
rather than installing where you didn't ask. To see what is installed and what
isn't across every harness × scope pair, run tome meta list — that is the
surface that shows up-to-date, stale, and not-installed side by side.
Pitfalls
- Exit
87— unknown skill id. The id you passed isn't a bundled meta skill.tome meta listshows which skills the binary includes. - Exit
88— install failed. Tome couldn't write the skill files — for example, it refuses to write through symlinked directories by design. The error says which target failed; installs into other harnesses still proceed. - Exit
89— no harness detected. Tome found no supported harness to install into. Pass--harness <name>explicitly, or run from a directory where a supported harness is set up.
The full table is in Exit codes.
Where next
- MCP server — agents can install meta skills themselves,
via the
metatool or the built-inadd-tome-conversion-skillprompt. - Converting — the commands the guided conversion runs.