fix(runtime): stop treating JSON bytes as hard token limit #467

Merged
karsten merged 2 commits from fix/model-input-false-hard-limit into main 2026-11-14 19:44:40 +00:00
Owner

Production 0.1.325 rejects legitimate Peter and Odo requests because serialized JSON byte counts are treated as tokens and hard-capped at 131072. Remove the false final admission gate while retaining compaction pressure and provider context limits; no history or effect pairs are truncated. Both working and summary calls can recover. Resource/cost boundary requires a future provider-aware token mechanism, not a byte=token comparison. Local typecheck/build and 2587 tests passed, 1 skipped; independent review PASS with P2 cost-boundary note. Existing operator work must not be replayed.

Production 0.1.325 rejects legitimate Peter and Odo requests because serialized JSON byte counts are treated as tokens and hard-capped at 131072. Remove the false final admission gate while retaining compaction pressure and provider context limits; no history or effect pairs are truncated. Both working and summary calls can recover. Resource/cost boundary requires a future provider-aware token mechanism, not a byte=token comparison. Local typecheck/build and 2587 tests passed, 1 skipped; independent review PASS with P2 cost-boundary note. Existing operator work must not be replayed.
fix(runtime): stop treating JSON bytes as hard token limit
All checks were successful
decision-interface-ci / full-validation (pull_request) Successful in 3m28s
8bccc2d77c
Author
Owner

PI-REVIEW-RECEIPT

State: PASS
Head: 8bccc2d77ca8
Summary: PASS: No blocking findings. The change removes coverage for an important oversized-tool-result failure path.
Findings: 1

Top findings:

  • P2 Replace the removed tool-effect recovery test (test/model-input-budget.test.ts)
<!-- PI-REVIEW-RECEIPT --> ## PI-REVIEW-RECEIPT State: **PASS** Head: `8bccc2d77ca8` Summary: PASS: No blocking findings. The change removes coverage for an important oversized-tool-result failure path. Findings: 1 Top findings: - **P2** Replace the removed tool-effect recovery test (test/model-input-budget.test.ts)
test: retain effect evidence when provider rejects large context
All checks were successful
decision-interface-ci / full-validation (pull_request) Successful in 3m28s
8a7eef392b
Author
Owner

PI-REVIEW-RECEIPT

State: PASS
Head: 8a7eef392b40
Summary: PASS: No blocking findings. The change leaves a non-blocking gap in compaction recovery coverage.
Findings: 1

Top findings:

  • P2 Retain coverage for failed summary requests after tool effects (test/model-provider-limit-recovery.test.ts:11)
<!-- PI-REVIEW-RECEIPT --> ## PI-REVIEW-RECEIPT State: **PASS** Head: `8a7eef392b40` Summary: PASS: No blocking findings. The change leaves a non-blocking gap in compaction recovery coverage. Findings: 1 Top findings: - **P2** Retain coverage for failed summary requests after tool effects (test/model-provider-limit-recovery.test.ts:11)
Sign in to join this conversation.
No reviewers
No labels
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference
harness/core!467
No description provided.