The Agent Skills Directory

[COMMAND_EXECUTION]: The skill dynamically generates and executes DataStage SDK code. In the fallback path of Step 5, it directly constructs sdk_code and invokes the create_datastage_flow tool to create new workflows. This is a core feature of the skill used for project optimization.
[DATA_EXPOSURE]: The workflow involves reading sensitive metadata, including project information, DataStage flow logic (sdk_code), and database connection UUIDs through the list_connections and retrieve_datastage_flow_code tools. This data access is scoped to the IBM/watsonx ecosystem and required for pushdown eligibility analysis.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface as it ingests existing DataStage flow code as untrusted data.
Ingestion points: Flow logic is ingested from retrieve_datastage_flow_code in SKILL.md Step 1.
Boundary markers: No explicit instruction delimiters or boundary markers are defined for the ingested code.
Capability inventory: The skill can create new flows and execute SQL through the create_datastage_flow tool and the di-agent-query-optimization skill.
Sanitization: The skill mitigates risks by using a structured mapping reference (references/stage-to-sql-mapping.md) to translate DataStage logic into SQL rather than performing open-ended interpretation.

di-agent-flow-pushdown-optimizer