The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes a local bash script, scripts/bridge-skill.sh, which performs file system operations such as creating directories and managing symbolic links. This execution is scoped to the skill's primary administrative function.
[PERSISTENCE]: The skill's core purpose is to establish persistence for other agent skills by creating symbolic links in standard agent configuration directories (e.g., ~/.claude/skills, ~/.agents/skills). This allows skills to remain available across different agent sessions and hosts.
[INDIRECT_PROMPT_INJECTION]: The skill interacts with user-specified directory paths and external skill metadata files, presenting an attack surface for indirect injection if malicious directories are targeted.
Ingestion points: The skill accepts directory names and paths from user input to determine the source for bridging operations.
Boundary markers: The implementation includes six hard rules in the documentation and enforces a requirement that target directories must contain a SKILL.md file.
Capability inventory: The bridge-skill.sh script utilizes standard shell commands (mkdir, ln, rm, cat) to modify the local file system within specific agent-related paths.
Sanitization: The script performs path resolution and utilizes a is_forbidden_source function to explicitly block directories named private, .private, or _internal, preventing the accidental or malicious exposure of sensitive folders.

xy-link