Best possible solution:

Rebase first, then either port only the retained broad projection behavior to the active current-main boundary with raw diagnostics and union semantics preserved, or leave the concrete Gemini fix to the merged provider-normalizer path and close/replace this draft by maintainer choice.

Do we have a high-confidence way to reproduce the issue?

Yes, source-level reproduction is high-confidence: the PR diff can hide raw dynamic-schema diagnostics, narrow mixed unions, and change ordered union coercion, while current main shows the safer active boundaries. I did not run a live provider request in this read-only review.

Is this the best way to solve the issue?

No as submitted. Provider schema compatibility is real, but this draft targets a stale boundary and changes shared contracts too broadly; the safer path is the current provider-normalizer fix for Gemini plus a separately rebased broad projection decision if maintainers still want it.

Full review comments:

• [P2] Run diagnostics before provider projection — src/agents/tool-schema-projection.ts:135
projectRuntimeToolInputSchema now returns the provider-normalized tree before dynamic-schema diagnostics run. A raw branch containing $dynamicRef or $dynamicAnchor can be collapsed or replaced before quarantine sees it, so keep diagnostics on the serialized raw schema and only project for provider output afterward.
  Confidence: 0.9
• [P2] Preserve mixed non-null union alternatives — src/agents/tool-schema-projection.ts:188-193
  The normalization falls back to the first string variant whenever a composition contains one, so string | number | null becomes only string. That misrepresents supported number inputs to providers and SDK consumers; dropping null branches must not discard other non-null alternatives.
  Confidence: 0.86
• [P2] Keep non-null union coercion ordered — packages/llm-core/src/validation.ts:207-212
  The exact-match pre-pass returns the original value for any matching union branch before ordered coercion. For anyOf [{type:"number"}, {type:"string"}], a numeric string now stays a string instead of following the existing first-branch numeric coercion; scope the shortcut to explicit null preservation.
  Confidence: 0.87

Overall correctness: patch is incorrect
Overall confidence: 0.9

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 843ad143647e.

Label justifications:

• P2: This is normal-priority agent/provider compatibility work with concrete blockers and a bounded provider-schema blast radius.
• merge-risk: 🚨 compatibility: The PR changes shared provider/Codex-visible tool schema projection and llm-core validation behavior used across provider and runtime paths.
• rating: 🧂 unranked krab: Overall readiness is 🧂 unranked krab; proof is 🌊 off-meta tidepool and patch quality is 🧂 unranked krab.
• status: ⏳ waiting on author: ClawSweeper has contributor-facing work open and is waiting for author action. Not applicable: The external-contributor proof gate is not applicable to this MEMBER-authored maintainer-labeled PR; the PR body provides targeted local validation but no live provider turn.

PR surface:

Source +83, Tests +109. Total +192 across 11 files.

What I checked:

• Repository policy applied: Root AGENTS.md was read fully, along with scoped src/agents, src/agents/tools, and test guides; provider/tool schema behavior is compatibility-sensitive and Codex-facing changes require direct sibling Codex source inspection. (AGENTS.md:14, 843ad143647e)
• Live PR state: GitHub reports this PR open, draft, maintainer-labeled, head daa3581, and mergeable=CONFLICTING / mergeStateStatus=DIRTY. (daa3581013fe)
• PR projects before raw diagnostics: The diff changes serialization to return normalizeProviderToolInputSchema(parsed), which means raw dynamic-schema diagnostics can inspect a provider-normalized tree instead of the original schema. (src/agents/tool-schema-projection.ts:135, daa3581013fe)
• Current main projection boundary: Current main makes src/agents/tool-schema-projection.ts a wrapper and keeps raw JSON projection and dynamic-keyword diagnostics in src/agents/tool-schema-json-projection.ts. (src/agents/tool-schema-projection.ts:1, 843ad143647e)
• Current main raw projection behavior: The active helper serializes, validates object shape, reports dynamic keywords, and returns projection.schema unchanged; broad provider cleanup is not in this helper on current main. (src/agents/tool-schema-json-projection.ts:115, 843ad143647e)
• Provider normalizer already owns the focused Gemini fix: Current main selects Gemini cleanup from provider id, Gemini model id, or toolSchemaProfile, and applies cleanSchemaForGemini in normalizeToolParameterSchemaUncached. (src/agents/agent-tools-parameter-schema.ts:758, 843ad143647e)

Likely related people:

• vincentkoc: Recent current-main history touches the tool schema helper exports, and this maintainer-authored draft plus the overlapping broader schema PR are in the same provider/runtime schema area. (role: recent area contributor; confidence: high; commits: e7e686db2d00, 361869e43422, daa3581013fe; files: src/agents/tool-schema-projection.ts, src/agents/tool-schema-json-projection.ts, packages/llm-core/src/validation.ts)
• Pick-cat: Authored the merged provider-normalizer fix that handles Gemini model ids behind OpenAI-compatible providers for the concrete cron schema rejection. (role: adjacent focused-fix author; confidence: medium; commits: 1f89d6d7f762; files: src/agents/agent-tools-parameter-schema.ts, src/agents/agent-tools.schema.test.ts, src/agents/tools/cron-tool.schema.test.ts)
• steipete: History shows substantial work on the provider schema normalizer and Gemini schema cleaner used by this path. (role: recent adjacent contributor; confidence: high; commits: 60e0d2a7b910, fe1fd055d557, 33aea44fe5b0; files: src/agents/agent-tools-parameter-schema.ts, src/agents/schema/clean-for-gemini.ts)
• openperf: Earlier merged cron/Gemini compatibility work removed OpenAPI 3.0-incompatible cron schema keywords in the same provider compatibility surface. (role: prior provider-schema contributor; confidence: medium; commits: 8c1ca1f24517; files: src/agents/schema/clean-for-gemini.ts, src/agents/tools/cron-tool.ts, src/agents/tools/cron-tool.schema.test.ts)

• 🦀 challenger crab: rare, exceptional readiness with strong proof, clean implementation, and convincing validation.
• 🦞 diamond lobster: very strong readiness with only minor maintainer review expected.
• 🐚 platinum hermit: good normal PR, likely mergeable with ordinary maintainer review.
• 🦐 gold shrimp: useful signal, but proof or patch confidence is still limited.
• 🦪 silver shellfish: thin signal; proof, validation, or implementation needs work.
• 🧂 unranked krab: not merge-ready because proof is missing/unusable or there are serious correctness or safety concerns.
• 🌊 off-meta tidepool: rating does not apply to this item.

Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

• ClawSweeper keeps one durable marker-backed review comment per issue or PR.
• Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
• A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
• PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
• Maintainers can also comment @clawsweeper review to request a fresh review only.
• Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
• Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
• Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.

• reviewed 2026-06-21T07:43:23.741Z sha daa3581 :: found issues before merge. :: [P2] Run diagnostics before provider projection | [P2] Preserve mixed non-null union alternatives | [P2] Keep non-null union coercion ordered