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.

Execution flow

1. Use idmp-cli element elements get --params and idmp-cli element elements path --params to lock the owner and business root.
2. Read dependencies with idmp-cli attribute elements attributes --params and idmp-cli analysis analyses list --params.
3. For natural-language requests, try AI draft-first create with idmp-cli ai create create --ack-risk --data '{"elementId":123,"prompt":"...","record":true,"deepThinking":false,"deviceDocument":false}', then persist the returned draft through idmp-cli analysis analyses create --ack-risk --params after removing id and setting rootElementId.
4. If the AI draft is unsuitable, persistence fails, or the backend returns a structured AI failure, fall back to the current structured path: reserve/validate with idmp-cli analysis analyses new-name --ack-risk --params, idmp-cli analysis trigger-types list --params, and idmp-cli attribute elements attributes-post --ack-risk --params if the output attribute does not exist.
5. Create and reread through idmp-cli analysis analyses create --ack-risk --params and idmp-cli analysis analyses get --params.
6. Finish with idmp-cli analysis analyses resume --ack-risk --params. Cleanup is best-effort: use idmp-cli analysis analyses delete --ack-risk --params only when the draft is abandoned or when the probe must be removed. If delete succeeds but the output attribute remains backend-referenced, record the leaked IDs and stop retrying generic delete loops.

Exception paths

• Stop when trigger types do not support the requested scope instead of forcing creation.
• If AI create returns an unusable draft, misses required scope details, or persistence fails, clean any draft-only output attributes and continue with the structured workflow instead of retrying the same draft blindly.
• If AI create times out, classify the first attempt as backend AI/API latency and move to the structured payload path without weakening the requested scope or runtime target.
• If reread shows the wrong rootElementId, delete the draft and rebuild the payload.
• Temporary output attributes should be deleted when the analysis draft is abandoned. If delete succeeds but the backend still reports those attributes as referred by the analysis, keep the IDs in the evidence and stop cleanup retries.
• Reread immediately and resume if needed.

Validation scenarios

1. Element-mode create on a leaf owner

Run idmp-cli analysis analyses create --ack-risk --params, then reread with idmp-cli analysis analyses get --params. In shared environments, once get and resume prove the analysis exists and runs, cleanup stays best-effort.

2. Element-mode create on a middle metric owner

Use idmp-cli analysis trigger-types list --params to prove the middle owner is metric-bearing. A plain container is not enough.

3. Element-mode hierarchy create with child template output

Keep child-scope creation gated by applyOnSelf=false trigger types. The payload still needs the child template in output.elementTemplate.id, but current live ELE_SUBET flows can reuse or create output attributes on the owner element instead of requiring attribute-template IDs.

4. Template-mode create with reused attribute template

If template mode is requested, preserve the candidate name and payload discipline. Do not collapse the payload to a guessed minimal DTO.

5. Root owner has no trigger types and the attempt must be cleaned up

If preflight fails on the chosen owner, stop before create. Any temporary output attribute created for the failed attempt must be removed, but if the backend leaks references after a proven create or delete, record the leak and stop retrying generic cleanup.