The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell commands to execute local Python scripts (normalize.py and isolation.py) from the plugin's directory. These scripts perform document processing and manage network state, representing standard vendor-provided functionality for this skill.
[EXTERNAL_DOWNLOADS]: The skill allows the agent to fetch data from external URLs. Safety protocols are mandated, including explicit user confirmation for domains and an immediate 'seal' of network access once the ingestion is complete.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it processes untrusted external content.
Ingestion points: Data is imported from local file systems or downloaded from the web into the <vault>/sources/ directory (SKILL.md).
Boundary markers: The instructions describe normalizing content into Markdown or CSV with front matter but do not include explicit commands to the agent to disregard instructions potentially embedded within the ingested content.
Capability inventory: The agent can run shell-based scripts via python3 to handle files and modify the system's network isolation settings.
Sanitization: There are no requirements for sanitizing or filtering the content of ingested files or downloaded pages before they are integrated into the vault library.

ingest