Skip to content
skilldropv0.1Documentation

Security model

Agent skill installation security

Understand how Skilldrop validates paths, rejects unsafe archive entries, verifies SHA-256 checksums, reports executables, and avoids automatic script execution.

Inspect before install

Agent skills are instructions and files that can influence powerful local tools. Skilldrop keeps review separate from installation with sk inspect, which verifies and displays a snapshot without extracting or executing it.

Archive validation

Before a snapshot is accepted or installed, Skilldrop validates the bundle and its manifest. The implementation rejects unsafe paths and link entries designed to escape the target directory.

  • Absolute paths and path traversal are rejected.
  • Symbolic and hard links are rejected.
  • Duplicate paths and files missing from the manifest are rejected.
  • Bundle and file SHA-256 checksums are verified.

Executable files

Inspection reports executable files separately so they cannot hide in a larger file tree. Skilldrop never automatically executes a script contained in a skill. You remain responsible for reviewing instructions before allowing an agent to use them.

Snapshot privacy and lifetime

Snapshot URLs are unlisted transfer links, not access-controlled secrets. Anyone with a URL can retrieve its contents. Do not upload credentials or material that must remain confidential.

Shared snapshots are immutable for exactly seven days, then deleted. A corrected share produces a new URL rather than changing the existing artifact. Snapshot URLs are unlisted, not private: anyone with the link can retrieve the contents before expiry.

Copied to clipboard