DOCS / WORKSPACES
Skills
The Skills explorer is a read-only browser for every skill your agents can discover:
your machine-level Claude skills, each repository’s .claude/skills, skills shipped
by installed plugins, and Codex’s skill folders. Use it to answer “what can this
agent actually do here?” without leaving Birch. (Background:
AI in Birch.) The feature is on by default and can be
toggled under Settings → Advanced → Skills explorer.
Three surfaces, one browser
Section titled “Three surfaces, one browser”- The Skills screen — the sidebar’s Skills item, showing every scope with scope and harness filters.
- The workspace “Repo skills” tab — scoped to that workspace’s worktree: exactly the project skills the agent running there can see. Filters are hidden; only project rows show.
- The properties-panel “Repo skills” button — a shortcut that opens the same tab.
What gets scanned
Section titled “What gets scanned”| Source | Scope | Notes |
|---|---|---|
~/.claude/skills | Machine (Claude) | One folder per skill, read from SKILL.md |
~/.claude/commands | Machine (Claude) | Legacy slash commands; rows carry a command badge |
.claude/skills in each repo | Project | From the repo checkout on the Skills screen; from the worktree in the Repo skills tab |
.claude/commands in each repo | Project | Legacy commands, same command badge |
| Installed Claude plugin skills | Plugin | Read from the plugin install records; plugins you’ve disabled in Claude Code are skipped |
~/.agents/skills | Machine (Codex) | Codex’s official skills root |
$CODEX_HOME/skills (default ~/.codex/skills) | Machine (Codex) | Legacy root; deduplicated by name against the official root, official wins |
Scanning is best-effort: a broken entry (commonly a dangling symlink into a cloud folder) skips only itself, never its siblings.
Filtering and refreshing
Section titled “Filtering and refreshing”The Skills screen offers a Filter skills search box, a scope filter (All / Machine / Project / Plugin), a harness filter (All / Claude / Codex), and a Refresh button. There is no file watcher — the list rescans when you enter the screen or tab, and on Refresh.
The detail pane
Section titled “The detail pane”Selecting a skill renders its SKILL.md with the frontmatter stripped. Remote images
are stripped for safety; local screenshots inside the skill folder still render.
Above the document sit the skill’s path and three actions:
- Reveal folder — open the skill’s folder in Finder / Explorer.
- Open in editor — open
SKILL.mdin your configured external editor. - Open in terminal — open a system terminal in the skill’s folder.
Read-only by design
Section titled “Read-only by design”Birch never edits, installs, or removes skills — it only shows you what’s on disk. To change a skill, use Open in editor and edit the file where it lives; the next rescan picks it up. Discovery is also not injection: listing a skill here doesn’t change what an agent loads — that’s still entirely between the agent CLI and its own skill folders.