The Agent Skills Directory

[REMOTE_CODE_EXECUTION] (CRITICAL): The commands alma update download and alma update install allow the agent to fetch and execute code from an unspecified remote source, enabling arbitrary software updates on the host system without verification.
[EXTERNAL_DOWNLOADS] (HIGH): The skill implements an update mechanism that retrieves remote assets from untrusted sources without integrity checks or source validation.
[CREDENTIALS_UNSAFE] (HIGH): The skill manages sensitive configuration values, including tts.apiKey. The alma config list and alma config get commands allow these credentials to be printed into the agent's context, where they can be observed or exfiltrated.
[PROMPT_INJECTION] (HIGH): The skill is vulnerable to Indirect Prompt Injection (Category 8) because SOUL.md is persistently injected into the system prompt. Ingestion points include the alma soul set command and direct Bash writes to ~/.config/alma/SOUL.md. There are no boundary markers or sanitization procedures, and the agent has high-privilege Bash and CLI capabilities, allowing malicious instructions to be stored in the agent's permanent identity.
[COMMAND_EXECUTION] (MEDIUM): The skill uses the Bash tool to perform arbitrary file manipulations and execute local CLI commands, providing a broad capability for system interaction.

self-management