Capability | What the skill does |
|---|
Search the OAC catalog | Searches OAC catalog items by name, including AI Agents, Workbooks, Datasets, Subject Areas, and Folders. Workbook is handled as the OAC project catalog type. |
Mapping of Agents with Datasets | Identifies dependencies between agents and datasets, or datasets columns |
Create an agent from a dataset | Resolves the dataset from the OAC catalog, inspects the dataset through MCP, samples rows through logical SQL, proposes a business scope, builds indexing rules, creates the agent, and optionally uploads document knowledge. |
Create an agent from a workbook | Parses an OAC workbook link or path, extracts the workbook datasource, selected columns, calculations, workbook filters, visual role bindings, parameters, and workbook usage patterns, then uses that context to propose a workbook-aware agent scope before creation. |
Handle Subject Area sources | Detects Subject Areas from direct requests or workbook datasources, inspects and samples them, helps refine the agent scope, then proposes scope-driven Logical SQL for a Local Subject Area before agent creation. LSA proposals default to 25 projected fields or fewer unless the user approves more. |
Inspect datasets before agent creation | Uses Oracle Analytics MCP to describe the dataset and run logical SQL samples, so the proposed agent scope, instructions, examples, and indexing are grounded in real metadata and sample rows. |
Manage dataset indexing independently | Updates dataset indexing without recreating the agent, using assistant-authored indexing payloads, saveDatasetProperties, qna/runNow, and qna/fetchLastRunIndex polling. |
Edit supplemental instructions independently | Fetches the current agent definition, changes only the targeted instruction fields, posts the full updated payload with overwrite enabled, and verifies the update afterward. |
Edit welcome messages independently | Updates the first-message or welcome-question content without rebuilding indexing or re-uploading document knowledge. |
Edit metadata independently | Updates agent purpose, description, history setting, and catalog description metadata while preserving existing datasource and document references. |
Manage document knowledge independently | Uploads, replaces, or deletes one document knowledge file while checking current agent documents and document metadata to avoid duplicates. |
Bulk edit instructions across agents | Supports bulk supplemental-instruction edits across multiple existing AI Agents when requested. This is handled as repeated targeted instruction updates: fetch each agent, change only the requested instruction text, save it back, and verify the update so unrelated fields are preserved. |
Generate few-shot document knowledge | Can generate a .txt document containing few-shot examples from the final question examples, upload it to the agent, and verify vectorization completion. |
Use provided document knowledge | Can upload a provided .txt or .pdf file under 5 MB as agent document knowledge, with verification after upload. |
Preserve existing indexing when requested | Detects existing dataset indexing rules and can create the agent while keeping current indexing unchanged if the user chooses that path. |
Override indexing when requested | Replaces existing indexing with new agent-specific indexing rules when the user chooses override or when no usable indexing exists. |
Work across OAC instances | Reads oac_url and tokens_path from config/oac-config.json so the same skill can operate against different OAC environments without hardcoded tenant URLs. |
