The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill is designed to ingest and parse untrusted local data from the filesystem, which could contain malicious instructions.
Ingestion points: Scans local directories (focus/, active/, paused/, etc.) for *.md files to extract metadata.
Boundary markers: Absent. The instructions do not define delimiters or provide 'ignore embedded instructions' warnings for the agent during the parsing process.
Capability inventory: Includes directory scanning, file reading, and data display. It lacks high-risk capabilities like file modification, command execution, or network access.
Sanitization: Absent. There is no mention of filtering or validating the content of the task headers beyond basic parsing.

task-list