Turn off per harness
fleet skill off tdd --harness pi disables one skill in one harness —
fleet skill on restores it.

Fleet keeps a single state file as the source of truth and projects it into each harness’s own config. Skills never move, the canonical store (~/.agents/skills) is never edited, and the skills CLI stays the install and update backend.
Turn off per harness
fleet skill off tdd --harness pi disables one skill in one harness —
fleet skill on restores it.
Interactive matrix
Bare fleet opens the skill × harness matrix: toggle, stage, and apply
together.
Custom skills
fleet skill adopt moves a custom skill into your skills collection and
links it into every harness.
Doctor & Sync
fleet skill doctor previews drift; fleet skill sync converges configs
with the state file.
| Harness | Per-skill off switch |
|---|---|
| OpenCode | ✅ |
| Pi | ✅ |
| Codex | ✅ |
| Claude Code | ✅ |
| Cursor | custom skills only |
| IBM Bob | custom skills only |
Cursor and Bob read the canonical store natively, so a stored skill has no per-skill off switch there. A custom skill does: it reaches them only through fleet’s managed symlink, so turning it off removes that link and turning it on restores it. See the per-harness reference for files touched, written shapes, and limitations.
| OS | Supported |
|---|---|
| macOS | ✅ |
| Linux | ✅ |
| Windows | ❌ |
skills by hand, resetting state, how sync decides