Update OpenAI API: use SDK v6 and replace legacy gpt-4-vision-preview model

[sepo-agent-app]

Summary

• Adds openai@^6.39.0 to package.json dependencies
• Replaces raw fetch calls in getHtmlFromOpenAI.ts, getCodeFromOpenAI.ts, and getInterpretationFromAI.ts with the official OpenAI SDK (client.chat.completions.create)
• Upgrades deprecated gpt-4-vision-preview → gpt-4o in getHtmlFromOpenAI.ts
• Fixes generateCode.tsx and interpretShapes.tsx to remove stale json.error checks (SDK throws on API errors) and handle null-safe content via ?? ''

Test plan

• npm install completes without dependency errors
• TypeScript build (npx tsc --noEmit) passes with no errors
• All three OpenAI functions use the SDK client with dangerouslyAllowBrowser: true
• No references to gpt-4-vision-preview remain in the codebase
• No raw fetch calls to api.openai.com remain in lib files

Closes #4

🤖 Generated with Claude Code

[ryanyen2]

@sepo-agent /orchestrate

[sepo-agent-app]

Sepo pushed fixes for this PR. Branch: agent/implement-issue-4/claude-26536075716. Requested by @sepo-agent.

• Removed unreachable if (!json) { throw Error('Could not contact OpenAI.') } dead-code blocks from app/lib/generateCode.tsx (lines 94–96) and app/lib/interpretShapes.tsx (lines 135–137) — the SDK always returns or throws, never null.
• Added a one-line intent comment above dangerouslyAllowBrowser: true in all three SDK client files (getHtmlFromOpenAI.ts, getCodeFromOpenAI.ts, getInterpretationFromAI.ts) clarifying the flag is intentional for this client-side demo app.

[sepo-agent-app]

Sepo is dispatching follow-up automation.

Source	Next	Target	Round	Status
fix-pr	review	PR #5	4 / 12	Dispatched

Reason: agent planner selected review: fix-pr succeeded on PR #5 (OpenAI SDK v6 migration). A review pass is warranted to verify the applied fixes are correct and the PR is ready for human review.

[sepo-agent-app]

No agent/rubrics branch exists in this repository (confirmed again). The previous rubric review in round 2 already returned N/A for the same reason.

Rubrics Review

Total Score	Verdict	Rubrics Scored
N/A	N/A	0

Dimension	Rubric	Result	Score	Evidence
—	—	not applicable	N/A	No agent/rubrics branch found for CodeShaping/annotate-edit; no user/team rubrics are configured

Notes

• No rubrics are configured for this repository. The agent/rubrics branch does not exist, so there is nothing to score against.
• This is round 4 of the orchestration loop (review pass after fix-pr). The prior fix-pr pass pushed changes to agent/implement-issue-4/claude-26536075716 addressing: removal of unreachable if (!json) dead-code blocks in generateCode.tsx and interpretShapes.tsx, and addition of intent comments at all three dangerouslyAllowBrowser: true instantiation sites.
• The PR diff confirms those changes are present in the current head (d62a5d197): dead-code guards removed, ?? '' null-coalescing applied, and intent comments added in at least getCodeFromOpenAI.ts. The PR is in a clean state for human review.
• To enable rubric scoring, initialize the agent/rubrics branch (e.g., via @sepo-agent /rubrics).

Findings

• INFO: No rubrics configured — scoring cannot proceed until the agent/rubrics branch is created.
• INFO (observation): fix-pr changes appear correctly applied. PR Update OpenAI API: use SDK v6 and replace legacy gpt-4-vision-preview model #5 now: removes bespoke fetch boilerplate (~200 lines), uses typed SDK v6 responses, upgrades gpt-4-vision-preview → gpt-4o, removes unreachable null guards, and documents the dangerouslyAllowBrowser intent. No new issues introduced.

Final Rubric Verdict

N/A

[sepo-agent-app]

AI Review Synthesis

Dual-agent review by Claude and Codex.
Requested by @ryanyen2.

Summary of PR/Issue

PR #5 migrates three hand-rolled fetch-based OpenAI wrappers (getCodeFromOpenAI.ts, getHtmlFromOpenAI.ts, getInterpretationFromAI.ts) from ~200 lines of bespoke boilerplate and locally re-invented type aliases to the official openai SDK v6 (openai@^6.39.0). It also upgrades the deprecated gpt-4-vision-preview model to gpt-4o. Two caller files (generateCode.tsx, interpretShapes.tsx) are updated to remove stale null-check guards and apply ?? '' null-coalescing on the SDK's string | null content field. The PR closes #4. A prior fix-pr pass (round 3) applied all mechanical fixes identified in the previous review cycle.

Review

All previously flagged WARNING-level issues have been resolved by the round-3 fix-pr agent. The current head (d62a5d197) is correct, complete, and ready for merge. Remaining findings are INFO-level, pre-existing, and non-blocking.

Issue	Severity	Description
No actionable issues remain	INFO	All WARNING findings (dead-code guards, missing dangerouslyAllowBrowser intent comments) confirmed fixed. Remaining observations are pre-existing or acceptable for this project's conventions.
Per-call OpenAI client instantiation	INFO	Each of the three wrapper functions creates a fresh client on every call. Acceptable because apiKey is user-supplied at call time and the SDK is stateless between calls; no blocker.
catch (e: any) wraps typed SDK errors	INFO	SDK error subtypes (APIError, RateLimitError, etc.) are collapsed to generic Error(e.message). Same behavior as prior raw-fetch implementation — not a regression, not a blocker.
getHtmlFromOpenAI.ts unreferenced	INFO	Pre-existing dead code; migrating it for consistency is the correct call. Noted for awareness only.
No automated tests	INFO	Test plan is fully manual; consistent with existing repo conventions. Not a blocker.

Progress

• Dead-code if (!json) { throw Error('Could not contact OpenAI.') } blocks removed from generateCode.tsx and interpretShapes.tsx — fixed in round-3 fix-pr pass.
• Dead-code if (json?.error) blocks removed — fixed in the original PR.
• ?? '' null-coalescing on message.content applied in both callers — correct, SDK types content as string | null.
• gpt-4-vision-preview fully replaced by gpt-4o in getHtmlFromOpenAI.ts.
• Intent comment added above dangerouslyAllowBrowser: true at all three SDK instantiation sites — fixed in round-3 fix-pr pass.
• Local hand-rolled type aliases (MessageContent, GPT4VCompletionRequest, GPT4oCompletionRequest) fully deleted.
• No rubrics configured for this repository (agent/rubrics branch does not exist) — rubric scoring is N/A.
• Inline thread cleanup this pass: Resolved two stale same-agent threads — PRRT_kwDOMNKfpc6FOMBP (app/lib/generateCode.tsx) and PRRT_kwDOMNKfpc6FOMC6 (app/lib/interpretShapes.tsx) — both confirmed fixed by the round-3 fix-pr commit.

Issue Details

No blocking or warning-level issues remain. All INFO findings are listed in the table above and require no branch changes.

Recommended Next Step

NO_AUTOMATED_ACTION — all concrete, automatable findings have been applied. The PR is ready for human review and merge.

Final Verdict

SHIP

Action Items

• (Optional) Consider initializing agent/rubrics branch to enable rubric scoring on future PRs (@sepo-agent /rubrics)