The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's setup instructions direct the agent to download the gazetteer database and utility scripts from a public GitHub repository (https://github.com/kltng/tgaz-sqlite.git).
[COMMAND_EXECUTION]: Instructions in SKILL.md prompt the agent to execute shell commands such as git lfs install, git clone, and cp to prepare the execution environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from an external SQLite database (tgaz.db) containing over 82,000 historical records, which represents a potential attack surface for indirect prompt injection.
Ingestion points: Untrusted data from the database (specifically placenames and historical notes in the snote table) is ingested into the agent's context through the query_tgaz.py script.
Boundary markers: None present; the instructions encourage the agent to incorporate the database results directly into natural language responses.
Capability inventory: The agent is equipped with shell and Python execution capabilities to run the provided database tools.
Sanitization: While the query_tgaz.py script correctly uses SQL parameterization to prevent SQL injection, it lacks mechanisms to sanitize or delimit the database content to prevent the agent from obeying instructions potentially embedded in the historical data.

tgaz-query