idmp-workflow-analysis-create

workflow: analysis create

Read ../idmp-shared/SKILL.md first.

Recommended references

• references/analysis-create.md
• ../idmp-analysis/SKILL.md

Missing context to resolve first

• Whether the request is natural-language friendly enough for AI-first create.
• Candidate analysis name.
• Runtime expectation.
• AI create prompt seed.
• idmp-cli ai create create --ack-risk --data '{"elementId":123,"prompt":"demo analysis prompt","record":true}'
• idmp-cli analysis analyses new-name --ack-risk --params '{"elementId":123,"name":"demo-analysis"}'
• idmp-cli analysis-template analyses new-name --ack-risk --params '{"elementTemplateId":456,"name":"demo-analysis"}'
• analysis.analyses.new-name and analysis-template.analyses.new-name require a proposed name value and --ack-risk.
• Live middle-owner proof plan.
• Whether the workflow is leaf self, middle self, or child aggregation.

Constrained live behaviors

• Prefer AI draft-first create for natural-language requests: POST /api/v1/ai/analysis/create first, then persist the returned draft through analysis analyses create.
• The AI draft request body follows ai.create.create: keep prompt, record, deepThinking, and deviceDocument explicit, plus either elementId or elementTemplateId.
• Minimal payloads fail in live environments.
• Create success does not guarantee Running.
• new-name for analyses requires a candidate name and --ack-risk.
• rootElementId is not the current element ID.
• A plain container plus ad-hoc attributes does not unlock self trigger types.
• element.elements.create and element.new.create do different jobs.
• Keep startAfterCreated, rootElementId, and output valueType explicit in the create payload.
• Trigger-type preflight decides whether the requested scope is valid before any create.
• AI create drafts can contain temporary output attributes or an id; remove the draft id, inject rootElementId, and clean draft-created attributes if persistence fails.
• If AI draft creation fails with a timeout such as context deadline exceeded, classify that first attempt as backend AI/API latency and fall back to the structured payload path without mutating the business intent.
• In the current live backend, analysis delete can return success while output attributes remain referenced. Treat cleanup after a proven create or running reread as best-effort instead of a hard create failure.