The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates dataset maintenance through specific tiangong-lca CLI commands. It includes built-in safeguards, requiring --commit and --confirm flags for dangerous operations and explicitly forbidding direct database manipulation or RLS bypass.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data from manifest and scope files. It implements a robust defense-in-depth strategy including ingestion point identification (manifest files), boundary verification through dry-runs, and mandatory readback verification to detect discrepancies.
[SAFE]: The skill enforces a strict security posture by limiting actions to the current user's authenticated scope and requiring multi-step verification (snapshotting, planning, and readback) for all operations.

dataset-rls-maintenance