The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands via the terminal tool for system configuration, such as setting up systemd user services and using sudo loginctl enable-linger to establish persistence for background synchronization tasks.
[COMMAND_EXECUTION]: The skill explicitly advises generating Python scripts in the /tmp/ directory and running them via the terminal tool to "avoid inline script security scanning entirely." This constitutes a deliberate technique to evade security monitoring mechanisms.
[EXTERNAL_DOWNLOADS]: The instructions recommend installing third-party software (obsidian-headless) from an external package registry and provide links to external GitHub repositories that are not identified as verified sources.
[PROMPT_INJECTION]: The skill creates an attack surface for indirect prompt injection by ingesting content from arbitrary URLs and documents (PDF, Word) without implementing boundary markers or "ignore embedded instructions" safeguards.
Ingestion points: web_extract for URLs, read_file for local document text.
Boundary markers: Absent.
Capability inventory: File system access, shell command execution, and dynamic Python execution.
Sanitization: Absent.
[CREDENTIALS_UNSAFE]: Instructions include managing environment variables in .env files and provide templates for login commands using placeholders for credentials. While placeholders are used, the workflow involves handling authentication data within the agent's execution environment.

llm-wiki