The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes subprocess.run within scripts/filebrowser_transfer/object_storage.py to execute a companion skill (object-storage). The path to the scripts for the companion skill is determined at runtime by searching through environment variables (OBJECT_STORAGE_SKILL_DIR), current and parent directories, and user home directories. This mechanism permits the execution of external code from locations resolved dynamically based on the environment.
[DYNAMIC_EXECUTION]: In scripts/filebrowser_transfer/object_storage.py, the CliObjectStorageGateway class constructs and executes shell commands using uv run. This pattern involves the runtime discovery and execution of external projects, which can lead to the execution of unintended code if the environment is misconfigured or compromised.
[INDIRECT_PROMPT_INJECTION]: The skill retrieves file names, metadata, and file content (when the --content flag is used) from remote FileBrowser instances and presents this information to the agent. Maliciously crafted data on the remote storage could attempt to influence the agent's behavior.
Ingestion points: The info, list_files, and search methods in scripts/filebrowser_transfer/filebrowser.py fetch data from the remote API.
Boundary markers: The skill outputs results in JSON format, but does not use specific delimiters or warnings to isolate this untrusted content within the agent's prompt.
Capability inventory: The skill has capabilities for network access (via httpx), file system writing (staging for transfers), and shell command execution (via subprocess.run).
Sanitization: While the skill implements path normalization to prevent directory traversal, it does not sanitize the contents of file metadata or bodies for embedded instructions before they are interpolated into the agent's context.

filebrowser