The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses system calls to interact with standard development and cloud tools.
Evidence: In scripts/optimization_applier.py, the subprocess.run function is used to execute databricks bundle commands for validation, deployment, and job execution.
Evidence: scripts/optimization_applier.py calls git diff to verify the state of the repository as part of its dual-persistence integrity check.
[PROMPT_INJECTION]: The skill has an attack surface for indirect prompt injection via optimized metadata proposals.
Ingestion points: The skill accepts candidate metadata and patch_set objects from external orchestrators as described in SKILL.md.
Boundary markers: There are no explicit delimiters or 'ignore' instructions implemented when interpolating these inputs into SQL comments or Genie Space instructions.
Capability inventory: The skill can execute DDL (e.g., ALTER TABLE, CREATE VIEW) and modify Genie Space configurations via the Databricks API.
Sanitization: Content is serialized using json.dumps for API delivery, ensuring structural safety, but the skill lacks content-level sanitization for the descriptive text injected into metadata fields.

genie-optimization-applier