The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/mailbox.py script executes the macOS security utility using subprocess.run to securely retrieve and store credentials in the system Keychain. The calls use specific arguments and avoid shell execution, reducing the risk of command injection. Additionally, the integration tests execute the mailbox script and OpenSSL via subprocess for environment setup and verification.
[DYNAMIC_EXECUTION]: The test files tests/mock_imap.py and tests/test_mailbox.py use the importlib module to load the mailbox.py script dynamically. This facilitates testing of the skill's logic using its own internal components and does not involve loading untrusted code from external sources.
[INDIRECT_PROMPT_INJECTION]: The skill processes external email subjects and bodies, which represents a surface for indirect prompt injection.
Ingestion points: Untrusted data enters the environment through IMAP fetch operations and is saved to the local MailKB directory in Markdown and JSON formats.
Boundary markers: Data is structured in Markdown files with YAML headers, providing clear separation between metadata and content, though the agent must still exercise caution when processing the email body content.
Capability inventory: The skill possesses the ability to write to the local file system, send outgoing emails via SMTP, and access the macOS Keychain for credential management.
Sanitization: The skill includes functions to strip HTML tags from email bodies and sanitize filenames to prevent directory traversal and script injection during the knowledge base construction process.

you-jian