feat: v2 - ai assistant - tangent researcher subagent

[maxy-shpfy]

Description

Adds a Tangent Researcher sub-agent to the Run View assistant that analyzes a pipeline run for ML optimization potential. The agent scores the run 0–100 and returns a prioritized list of hyperparameter-tuning and experiment ideas across four categories: feature engineering, hyperparameter optimization, input data, and model architecture.

The agent is read-only — it calls get_run_status and get_pipeline_state but cannot edit the pipeline or submit runs. It runs with high reasoning effort and is exposed to the Run View dispatcher as the create_optimization_scenario tool.

The response is returned as a fenced ```tangent-scenario code block containing a structured JSON payload. A new TangentScenario React component parses this block and renders the score, rationale, summary, and each idea as an interactive card with impact and type badges. The dispatcher prompt instructs the orchestrator to relay this block verbatim so the UI can render it correctly.

Related Issue and Pull requests

Type of Change

• Bug fix
• New feature
• Improvement
• Cleanup/Refactor
• Breaking change
• Documentation update

Checklist

• I have tested this does not break current pipelines / runs functionality
• I have tested the changes on staging

Screenshots (if applicable)

Test Instructions

1. Open a pipeline run in the Run View.
2. Ask the assistant something like "How can I optimize this run?" or "What hyperparameters should I tune for run 12345?".
3. Confirm the dispatcher routes to create_optimization_scenario rather than ask_debug_assistant or ask_general_help.
4. Verify the response renders as a score badge, rationale, summary paragraphs, and idea cards — not raw JSON or a plain code block.
5. Confirm that asking debug or general questions still routes to the correct specialist and is unaffected.

Additional Comments

The TangentScenario component degrades gracefully — if the JSON cannot be parsed, it falls back to rendering the raw string as plain text.

[github-actions]

🎩 Preview

A preview build has been created at: 06-02-feat_v2_-_ai_assistant_-_tangent_researcher_subagent/145b581

[maxy-shpfy]

Warning

This pull request is not mergeable via GitHub because a downstack PR is open. Once all requirements are satisfied, merge this PR as a stack on Graphite.
Learn more

• feat: v2 - tangent - connect dashboard to experiment scenarios #2387
• feat: v2 - tangent - progress update poll #2386
• feat: v2 - tangent - persist url to research session in opencode agent #2371
• feat: v2 - tangent - facelift for run details panel #2369
• feat: v2 - tangent - suggest optimization button #2368
• feat: v2 - tangent - optimization scenario card condensed #2367
• feat: v2 - tangent - autoresearch run #2363
• feat: v2 - tangent - scenarios window #2362
• feat: v2 - ai assistant - tangent researcher subagent #2361 👈 (View in Graphite)
• poc: scaffold basic tangent dashboard ux #2360
• chore: generate api sdk from tangent #2359
• refactor: v2 - run view - run menu #2358
• refactor: v2 - run view - windows top menu #2357
• fix: v2 - tangent - cors worker load #2364
• master

This stack of pull requests is managed by Graphite. Learn more about stacking.