The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill manages Ethereum private keys to establish agent identity. It implements several security measures:
Support for memory-only storage via environment variables (BASEMAIL_PRIVATE_KEY).
Local storage in ~/.basemail/ using AES-256-GCM encryption (scripts/crypto-utils.js).
Enforces strict file permissions (0o600 for keys/tokens and 0o700 for the configuration directory).
Warns users against committing keys to version control and provides a .gitignore template.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the api.basemail.ai inbox service.
Ingestion points: scripts/inbox.js reads data.body and data.snippet from the remote API.
Boundary markers: None identified; email content is logged directly to the console without encapsulation.
Capability inventory: The agent can perform network operations (sending emails via scripts/send.js) and execute internal scripts (scripts/register.js via spawnSync).
Sanitization: The skill does not sanitize or escape email content before displaying it, creating a surface where malicious instructions in an email could influence agent logic.
[COMMAND_EXECUTION]: scripts/inbox.js and scripts/send.js use child_process.spawnSync to execute node scripts/register.js for automatic token re-issuance. This is a functional requirement for the skill's operation and targets a known internal path.

BaseMail