The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses a Bash script (scripts/fetch-metadata.sh) and multiple shell commands (tar, find, rm, mkdir) to manage metadata archives downloaded from the network.
[PROMPT_INJECTION]: An indirect prompt injection surface exists as the agent is instructed to read and process YAML files fetched from a remote Metabase instance to understand the data model.
Ingestion points: YAML metadata files stored in the .metadata_cache/databases/ directory after being fetched from the remote Metabase URL.
Boundary markers: Absent. The skill does not define delimiters or instructions for the agent to ignore potentially malicious instructions embedded in the metadata.
Capability inventory: The agent has access to Bash, Write, Edit, and WebFetch tools, which could be exploited if malicious content is processed.
Sanitization: No validation or sanitization is performed on the remote YAML content before it enters the agent's context.
[COMMAND_EXECUTION]: The SKILL.md file contains the ! command syntax (! bash ...) which can trigger shell execution at the moment the skill is loaded on supported platforms.
[EXTERNAL_DOWNLOADS]: The skill facilitates the download of a compressed archive (.tar.gz) from a user-provided Metabase URL, which is then extracted and processed locally.

metabase-database-metadata