Skip to content

Keep onboarding to one model attempt and Tool counts to one turn - #1002

Merged
Y1fe1Zh0u merged 1 commit into
mainfrom
fix/v1.11.4-onboarding-turn-tool-count
Aug 24, 2026
Merged

Keep onboarding to one model attempt and Tool counts to one turn#1002
Y1fe1Zh0u merged 1 commit into
mainfrom
fix/v1.11.4-onboarding-turn-tool-count

Conversation

@Y1fe1Zh0u

Copy link
Copy Markdown
Collaborator

Summary

  • Restrict synthetic onboarding to one primary model attempt: no provider retry, fallback, wait Tool, model repair loop, or semantic Completion Gate repair.
  • Preserve deterministic completion checks and normal terminal delivery so the streaming bubble and Stop control settle from Runtime truth.
  • Display each AnalysisCard Tool count from the current user-turn group instead of the Session-wide aggregate.

Verification

  • Backend Runtime/onboarding/WebSocket: 179 passed
  • Frontend: 122 passed
  • Frontend production build: passed
  • Scoped Ruff: passed
  • Architecture Guard P0: passed
  • git diff --check: passed

Release

  • After merge, recreate v1.11.4-fix.1 from the resulting main commit. The prior GitHub Release is deleted; the old remote tag will be removed only after this PR is merged.

Synthetic onboarding now makes one primary provider attempt, exposes no wait tool, skips fallback and model-authored repair, and bypasses the semantic completion gate after deterministic checks pass. Direct Chat analysis cards count only Tool rows inside their user-turn group instead of substituting the Session aggregate.

Constraint: Ordinary chat Runs retain their existing retry, fallback, repair, and semantic completion behavior.

Rejected: Hide the stop button while onboarding remains active | would conceal a genuinely cancellable durable Run without fixing its repeated work.

Rejected: Recompute per-turn counts in the backend Session list | the frontend already loads complete folded Tool turns and owns the user-turn grouping boundary.

Confidence: high

Scope-risk: moderate

Reversibility: clean

Directive: Keep onboarding_target_phase as the explicit boundary for single-attempt behavior; do not infer it from empty visible history.

Tested: 179 backend tests; 122 frontend tests; frontend production build; scoped Ruff; Architecture Guard P0; git diff --check.

Not-tested: Live onboarding against the production provider and browser.
@Y1fe1Zh0u
Y1fe1Zh0u merged commit 920631d into main Aug 24, 2026
1 check was pending
@Y1fe1Zh0u
Y1fe1Zh0u deleted the fix/v1.11.4-onboarding-turn-tool-count branch August 24, 2026 12:04
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant