fix(openai): unwrap SELECTION-type setting to a string — TRUE root cause of HTTP 400

[ComBba]

TRUE ROOT CAUSE

settings.get('openaiModel') on a Devvit SELECTION-type field returns a string array (["gpt-5.4-mini"]), not a string. We were passing the raw array straight into the request body as "model": ["gpt-5.4-mini"]. OpenAI rejected this with the misleading error We could not parse the JSON body — which sounds like structural JSON error but is actually a type mismatch on the model field.

Production proof (v0.0.39 logs)

[vibe-mod] callOpenAI: openaiModel raw = ["gpt-5.4-mini"]

This single bug explains why PR #32-#37 (six rounds of unrelated body-shape and content fixes) all still returned HTTP 400. None of them touched the model field. PR #38 worked around it by hardcoding gpt-5.4-nano (a string literal), at which point callOpenAI started succeeding — but our few-shot was still in the broken flattened format from PR #37, so v0.0.39 returned a different toast: The compiled rule contained an action this app does not support. That's a downstream issue, not the transit bug.

Why probes (b)/(d)/(e)/(f) all returned 200

All probes hardcoded 'gpt-5.4-nano' (a string literal) and so sent valid "model": "gpt-5.4-nano" to OpenAI. Only callOpenAI read through settings.get(...) and got the array-of-string-instead-of-string mismatch.

Fix

1. Unwrap the SELECTION result: if Array.isArray(raw), take raw[0].
2. Log both the raw and unwrapped values for diagnostic visibility.
3. Honour the user-configured model (default gpt-5.4-mini).
4. Restore inline JSON.stringify(ex.assistant) for each few-shot example — re-teaches the model the correct {action, params} shape that the executor validates.

Tests

• npm run check 4/4 PASS.
• After merge: devvit upload → v0.0.40 → Chrome verify → production callOpenAI returns 200 → toast shows compile success.

Postmortem note

The misleading OpenAI error wording ("could not parse the JSON body" for a type mismatch on a single field, rather than e.g. "Invalid type for parameter 'model'") cost us seven PRs and ~40 minutes of bisecting. We tried body shape, escape density, byte vs string, message count, size, Content-Length, JSON-syntax-in-content — none addressed the actual issue. The diagnostic console.log added in PR #38 ("settings.get(openaiModel) = ...") surfaced the array in one line.

🤖 Generated with Claude Code

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.

[coderabbitai]

Warning

Rate limit exceeded

@ComBba has exceeded the limit for the number of commits that can be reviewed per hour. Please wait 43 minutes and 6 seconds before requesting another review.

You’ve run out of usage credits. Purchase more in the billing tab.

⌛ How to resolve this issue?

After the wait time has elapsed, a review can be triggered using the @coderabbitai review command as a PR comment. Alternatively, push new commits to this PR.

We recommend that you space out your commits to avoid hitting the rate limit.

🚦 How do rate limits work?

CodeRabbit enforces hourly rate limits for each developer per organization.

Our paid plans have higher rate limits than the trial, open-source and free plans. In all cases, we re-allow further reviews after a brief timeout.

Please see our FAQ for further information.

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: f1918fd2-059d-4a94-8f1a-53341c74012b

📥 Commits

Reviewing files that changed from the base of the PR and between b2cc1c1 and 6a59ea2.

📒 Files selected for processing (1)

• src/server/index.ts

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch fix/openai-400-unwrap-selection

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[gemini-code-assist]

Code Review

This pull request fixes a bug where the openaiModel setting was incorrectly being sent as an array instead of a string, and restores the use of structured JSON in few-shot examples now that the model-array issue is resolved. The reviewer suggests that since the previous 'JSON-syntax-free' strategy has been abandoned, the prompt construction should move away from single-line collapsing and instead use newlines to better delineate sections, providing a code suggestion to implement this improved structure.

[gemini-code-assist]

Now that the 'Round-6 strategy' of removing all JSON syntax has been abandoned to restore proper JSON few-shot examples (line 1313), the use of collapse and join(' ') to create a single-line prompt is less justified. While the author mentions a 'transit-safe' shape, the restoration of JSON.stringify already introduces significant escape sequences (\", {, }, etc.).

Using newlines and preserving the structure of the system prompt would likely improve the model's ability to follow instructions and distinguish between sections (Instructions, Examples, Task Input). Consider using \n as a separator and removing the collapse calls on the system prompt and user input.

Suggested change

	const collapse = (s: string) => s.replace(/\s+/g, ' ').trim();
	// Recursively flatten any value to a JSON-syntax-free string. Keys use
	// `key=value`, arrays use `[a; b; c]`-but-with-no-brackets style, objects
	// are space-separated `key=value` pairs.
	const flattenValue = (v: unknown): string => {
	if (v == null) return 'null';
	if (typeof v === 'string') return v.replace(/\s+/g, ' ');
	if (typeof v === 'number' || typeof v === 'boolean') return String(v);
	if (Array.isArray(v)) return v.map(flattenValue).join(' or ');
	if (typeof v === 'object') {
	return Object.entries(v as Record<string, unknown>)
	.map(([k, val]) => `${k}=${flattenValue(val)}`)
	.join(' ');
	}
	return String(v);
	};
	const flattenExample = (ex: (typeof FEW_SHOT_EXAMPLES)[number]): string => {
	const obj = ex.assistant as Record<string, unknown>;
	const flat = Object.entries(obj)
	.map(([k, v]) => `${k}=${flattenValue(v)}`)
	.join('; ');
	return `EXAMPLE INPUT ${collapse(ex.user)} EXAMPLE OUTPUT ${flat}`;
	};
	const SHORT_PROMPT_CHARS = 3500;
	const systemShort = collapse(VIBE_MOD_SYSTEM_PROMPT).slice(0, SHORT_PROMPT_CHARS);
	const firstExample = FEW_SHOT_EXAMPLES[0];
	const exampleText = firstExample ? flattenExample(firstExample) : '';
	const exampleBlock = (ex: (typeof FEW_SHOT_EXAMPLES)[number], i: number): string =>
	`EXAMPLE ${i + 1} INPUT: ${collapse(ex.user)} EXAMPLE ${i + 1} OUTPUT: ${JSON.stringify(ex.assistant)}`;
	const compositeContent = [
	'SYSTEM INSTRUCTIONS',
	systemShort,
	exampleText,
	'TASK INPUT',
	'SYSTEM INSTRUCTIONS:',
	collapse(VIBE_MOD_SYSTEM_PROMPT),
	'EXAMPLES:',
	...FEW_SHOT_EXAMPLES.map(exampleBlock),
	'TASK INPUT:',
	collapse(userRule.slice(0, 1000)),
	...(clarif ? ['TASK CLARIFICATION', collapse(clarif)] : []),
	'OUTPUT strict JSON only no prose',
	]
	.filter(Boolean)
	.join(' ');
	...(clarif ? ['TASK CLARIFICATION:', collapse(clarif)] : []),
	'OUTPUT (strict JSON object only, no prose):',
	].join(' ');
	const exampleBlock = (ex: (typeof FEW_SHOT_EXAMPLES)[number], i: number): string =>
	`EXAMPLE ${i + 1} INPUT: ${ex.user.trim()}\nEXAMPLE ${i + 1} OUTPUT: ${JSON.stringify(ex.assistant)}`;
	const compositeContent = [
	'SYSTEM INSTRUCTIONS:',
	VIBE_MOD_SYSTEM_PROMPT.trim(),
	'',
	...(FEW_SHOT_EXAMPLES.length > 0 ? ['EXAMPLES:', ...FEW_SHOT_EXAMPLES.map(exampleBlock), ''] : []),
	'TASK INPUT:',
	userRule.slice(0, 1000).trim(),
	...(clarif ? ['', 'TASK CLARIFICATION:', clarif.trim()] : []),
	'',
	'OUTPUT (strict JSON object only, no prose):',
	].join('\n');