The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents interfaces for processing external data from the web and image analysis, which introduces a potential vector for indirect prompt injection.
Ingestion points: In SKILL.md and references/llms-txt.md, the skill provides documentation for the web search API (/api/web_search) and vision model interactions which ingest untrusted external data.
Boundary markers: The documented examples do not implement boundary markers or instructions to ignore embedded commands within the external data.
Capability inventory: The skill documents extensive system interaction capabilities, including the ollama CLI and shell commands via curl.
Sanitization: There is no documentation regarding the sanitization or validation of data retrieved from external sources before it is processed by the agent.
[COMMAND_EXECUTION]: The skill provides comprehensive examples of shell commands for interacting with local services and managing AI models.
Evidence: SKILL.md and references/llms-txt.md contain numerous CLI examples using curl, ollama, and environment variable management tools.
[PRIVILEGE_ESCALATION]: The provided documentation includes administrative commands for system service management and uninstallation cleanup.
Evidence: references/llms-txt.md documents the use of sudo for tasks such as changing directory ownership (sudo chown) and removing system libraries (sudo rm -rf /usr/lib/ollama).
[EXTERNAL_DOWNLOADS]: The documentation provides instructions for downloading the Ollama installation script.
Evidence: references/llms-txt.md references re-running the install script from the official vendor domain.
[REMOTE_CODE_EXECUTION]: The skill documents the standard installation procedure for Ollama which involves piping a remote script to the shell.
Evidence: references/llms-txt.md refers to the vendor's installation script which is typically executed via curl | sh.

ollama