The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external JSON manifests to determine the scope of maintenance tasks.
Ingestion points: The skill reads task definitions and target scopes from maintenance-scope.json and redo-source-manifest.json as specified in the workflow section of SKILL.md.
Boundary markers: The skill includes explicit Boundaries and Stop Conditions sections that prevent operations on data not visible under current-user RLS.
Capability inventory: The skill coordinates row-level deletion, retirement, and redo-import operations across multiple dataset tables using CLI commands.
Sanitization: The workflow implements a mandatory human-in-the-loop approval process (approval-record.json) and post-commit readback verification (readback-verify-report.json) to mitigate injection risks.
[COMMAND_EXECUTION]: The skill utilizes the tiangong-lca CLI for all database and authentication operations.
Evidence: Invokes command families such as tiangong-lca dataset maintenance and tiangong-lca auth. These commands are identified as vendor-specific resources provided by the skill author.

dataset-rls-maintenance