Expand description
Install the embedded skills into agent skill directories, as packages.
The destinations live outside any instance — under the invoking user’s
home — so nothing here touches an instance manifest. The unit is the
package: one directory per skill, holding SKILL.md and every shared
artifact under references/, which is what the Agent Skills format and
every documented host resolve against.
Three references decide every destination: the payload, the receipt, and
the disk. Bytes matching the payload are current. Bytes the receipt
vouches for are this tool’s and may be replaced or taken back. Every
other byte is the user’s: an install refuses it without --force, and
an uninstall leaves it and names it.
Every apply holds the user-scope lock for its whole run, stages each write beside its destination, replaces one file at a time, and writes the receipt last. A run the process does not finish leaves whole files and the previous receipt, reports what it wrote, and is safe to run again.
Re-exports§
pub use crate::domain::paths::AGENTS_ROOT;pub use crate::domain::paths::CLAUDE_ROOT;pub use crate::domain::paths::LEGACY_SHARED_ROOT;
Structs§
- Layout
- Where one run writes: the agent roots, the state root, and the receipt.