The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external data that could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: The skill reads HTML files from the user's ~/Downloads directory (Magic: The Gathering Reddit threads processed via thread-extract) and fetches premium card statistics from MTG Arena sites (premium.json).
Boundary markers: The skill contains internal rules for the agent to follow (e.g., 'Iron Rules'), but lacks explicit delimiters or instructions to ignore potential commands embedded in the ingested HTML or JSON data.
Capability inventory: The agent has the ability to write files (via the Artifact tool), execute shell commands (twohg-scan, limited-stats, etc.), and perform network operations.
Sanitization: No explicit sanitization or filtering is described for the external content before it is processed or included in the final artifact.
[COMMAND_EXECUTION]: The workflow relies on executing several local CLI tools to process card data and player reports. These tools are part of a shared mtg_utils package provided by the author.
[EXTERNAL_DOWNLOADS]: The skill downloads card data and rulebooks from well-known Magic: The Gathering data services and repositories. These operations are essential for the skill's stated purpose.

twohg-guide