The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides a command to update itself by fetching and executing a shell script (install.sh) from the vendor's official GitHub repository. As this resource is provided by the identified author 'Canner', it represents standard vendor functionality.
[EXTERNAL_DOWNLOADS]: The skill performs a network request to the vendor's GitHub repository to retrieve version information (versions.json) for update checking purposes.
[PROMPT_INJECTION]: The skill ingests untrusted data from an external database to build manifests, presenting an indirect prompt injection surface.
Ingestion points: Metadata is ingested via the list_remote_tables() and list_remote_constraints() tools as described in SKILL.md.
Boundary markers: No explicit delimiters or instructions are used to separate the ingested metadata from the agent's instructions during the MDL construction process.
Capability inventory: The skill has the capability to deploy manifests and perform dry runs via deploy_manifest() and dry_run() tools.
Sanitization: There is no mention of sanitizing or escaping column/table names or constraints before they are interpolated into the manifest structure.
[SAFE]: Sensitive database credentials are never handled directly by the skill prompts or the ibis-server API; instead, users are directed to manage connections through a local Web UI or a local JSON file.

generate-mdl