Release v1.0.2 - Smart caching and stability improvements

## 🎯 Major Improvements ### Smart ETag-Based Caching - Replaced 24-hour TTL cache with HTTP ETag-based conditional requests - Only downloads instructions when content actually changes (304 Not Modified responses) - Significantly reduces GitHub API calls while staying up-to-date ### Release Tag Tracking for Stability - Now fetches Codex instructions from latest GitHub release tag instead of main branch - Ensures compatibility with ChatGPT Codex API (main branch may have unreleased features) - Prevents "Instructions are not valid" errors from bleeding-edge changes ## 🐛 Bug Fixes ### Model Normalization - Fixed default model fallback: unsupported models now default to `gpt-5` (not `gpt-5-codex`) - Preserves user's choice between `gpt-5` and `gpt-5-codex` when explicitly specified - Only codex model variants normalize to `gpt-5-codex` ### Error Prevention - Added `body.text` initialization check to prevent TypeError on `body.text.verbosity` - Improved error handling in request transformation ### Code Quality - Standardized all console.error prefixes to `[openai-codex-plugin]` - Updated documentation to reflect ETag caching implementation - Added npm version and downloads badges to README ## 📚 Documentation - Updated README with accurate caching behavior description - Added npm package badges for version tracking - Clarified release-based fetching strategy Co-authored-by: Hanzo Dev <[email protected]>
2026-09-24 17:01:02 +00:00 · 2025-10-02 12:13:46 +01:00
parent 12348a68af
commit a6e52abac3
6 changed files with 91 additions and 44 deletions
@@ -1,5 +1,8 @@
 # OpenAI ChatGPT OAuth Plugin for opencode

+[![npm version](https://img.shields.io/npm/v/opencode-openai-codex-auth.svg)](https://www.npmjs.com/package/opencode-openai-codex-auth)
+[![npm downloads](https://img.shields.io/npm/dm/opencode-openai-codex-auth.svg)](https://www.npmjs.com/package/opencode-openai-codex-auth)
+
 This plugin enables opencode to use OpenAI's Codex backend via ChatGPT Plus/Pro OAuth authentication, allowing you to use your ChatGPT subscription instead of OpenAI Platform API credits.

 ## Features
@@ -7,7 +10,7 @@ This plugin enables opencode to use OpenAI's Codex backend via ChatGPT Plus/Pro
 - ✅ ChatGPT Plus/Pro OAuth authentication
 - ✅ **Zero external dependencies** - Lightweight with only @openauthjs/openauth
 - ✅ **Auto-refreshing tokens** - Handles token expiration automatically
- ✅ **Auto-updating Codex instructions** - Fetches latest from OpenAI's Codex repo (cached 24h)
+- ✅ **Smart auto-updating Codex instructions** - Tracks latest stable release with ETag caching
 - ✅ Full tool support (write, edit, bash, grep, etc.)
 - ✅ Automatic tool remapping (Codex tools → opencode tools)
 - ✅ High reasoning effort with detailed thinking blocks
@@ -87,9 +90,12 @@ The plugin automatically configures:
 The plugin:

 1. **Authentication**: Uses ChatGPT OAuth flow with PKCE for secure authentication
-2. **Token Management**: Auto-refreshes tokens using `ai-sdk-provider-chatgpt-oauth`
-3. **Codex Instructions**: Automatically fetches latest instructions from [openai/codex](https://github.com/openai/codex) repository
-   - Cached locally in `~/.opencode/cache/` for 24 hours
+2. **Token Management**: Native token refresh implementation (no external dependencies)
+3. **Codex Instructions**: Automatically fetches from the latest stable release of [openai/codex](https://github.com/openai/codex)
+   - Tracks latest release tag (not main branch) for stability
+   - Uses ETag-based caching for efficient updates (only downloads when content changes)
+   - Cached locally in `~/.opencode/cache/`
+   - Auto-updates when OpenAI publishes new releases
   - Falls back to bundled version if GitHub is unavailable
 4. **Request Transformation**: Routes requests to `https://chatgpt.com/backend-api/codex/responses`
 5. **Model Normalization**: Maps all model names to `gpt-5-codex` (the Codex backend model)
@@ -40,7 +40,7 @@ export async function OpenAIAuthPlugin({ client }) {
 					return {};
 				}

-				// Fetch Codex instructions (cached for 24h)
+				// Fetch Codex instructions (cached with ETag)
 				const CODEX_INSTRUCTIONS = await getCodexInstructions();

 				// Return options that will be passed to the OpenAI SDK
@@ -100,21 +100,17 @@ export async function OpenAIAuthPlugin({ client }) {
 							try {
 								const body = JSON.parse(init.body);

-								// Normalize model name - Codex only supports specific model IDs
-								// Map all variants to their base models
+								// Normalize model name - Codex only supports gpt-5 and gpt-5-codex
 								if (body.model) {
 									if (body.model.includes("codex")) {
 										// Any codex variant → gpt-5-codex
 										body.model = "gpt-5-codex";
-									} else if (
-										body.model.includes("gpt-5") ||
-										body.model.includes("gpt-nano")
-									) {
-										// gpt-5 variants → gpt-5-codex for best tool support
-										body.model = "gpt-5-codex";
+									} else if (body.model.includes("gpt-5")) {
+										// gpt-5 variants → gpt-5 (keep user's choice)
+										body.model = "gpt-5";
 									} else {
-										// Default fallback
-										body.model = "gpt-5-codex";
+										// Default fallback for unsupported models
+										body.model = "gpt-5";
 									}
 								}

@@ -153,6 +149,9 @@ export async function OpenAIAuthPlugin({ client }) {
 								}
 								body.reasoning.effort = "high";
 								body.reasoning.summary = "detailed";
+								if (!body.text) {
+									body.text = {};
+								}
 								body.text.verbosity = "medium";

 								// Remove unsupported parameters
@@ -72,7 +72,7 @@ export async function exchangeAuthorizationCode(
 	});
 	if (!res.ok) {
 		const text = await res.text().catch(() => "");
-		console.error("[openai-codex-auth] code->token failed:", res.status, text);
+		console.error("[openai-codex-plugin] code->token failed:", res.status, text);
 		return { type: "failed" };
 	}
 	const json = await res.json();
@@ -81,7 +81,7 @@ export async function exchangeAuthorizationCode(
 		!json?.refresh_token ||
 		typeof json?.expires_in !== "number"
 	) {
-		console.error("[openai-codex-auth] token response missing fields:", json);
+		console.error("[openai-codex-plugin] token response missing fields:", json);
 		return { type: "failed" };
 	}
 	return {
@@ -4,61 +4,103 @@ import { fileURLToPath } from "node:url";
 import { homedir } from "node:os";

 // Codex instructions constants
-const CODEX_INSTRUCTIONS_URL =
-	"https://raw.githubusercontent.com/openai/codex/main/codex-rs/core/gpt_5_codex_prompt.md";
+const GITHUB_API_RELEASES = "https://api.github.com/repos/openai/codex/releases/latest";
 const CACHE_DIR = join(homedir(), ".opencode", "cache");
 const CACHE_FILE = join(CACHE_DIR, "codex-instructions.md");
 const CACHE_METADATA_FILE = join(CACHE_DIR, "codex-instructions-meta.json");
-const CACHE_TTL = 24 * 60 * 60 * 1000; // 24 hours

 const __filename = fileURLToPath(import.meta.url);
 const __dirname = dirname(__filename);

 /**
- * Fetch Codex instructions from GitHub and cache them
+ * Get the latest release tag from GitHub
+ * @returns {Promise<string>} Release tag name (e.g., "rust-v0.43.0")
+ */
+async function getLatestReleaseTag() {
+	const response = await fetch(GITHUB_API_RELEASES);
+	if (!response.ok) throw new Error(`Failed to fetch latest release: ${response.status}`);
+	const data = await response.json();
+	return data.tag_name;
+}
+
+/**
+ * Fetch Codex instructions from GitHub with ETag-based caching
+ * Uses HTTP conditional requests to efficiently check for updates
+ * Always fetches from the latest release tag, not main branch
 * @returns {Promise<string>} Codex instructions
 */
 export async function getCodexInstructions() {
 	try {
-		// Check if cache exists and is fresh
+		// Get the latest release tag
+		const latestTag = await getLatestReleaseTag();
+		const CODEX_INSTRUCTIONS_URL = `https://raw.githubusercontent.com/openai/codex/${latestTag}/codex-rs/core/gpt_5_codex_prompt.md`;
+
+		// Load cached metadata (includes ETag and tag)
+		let cachedETag = null;
+		let cachedTag = null;
 		if (existsSync(CACHE_METADATA_FILE)) {
 			const metadata = JSON.parse(readFileSync(CACHE_METADATA_FILE, "utf8"));
-			const age = Date.now() - metadata.timestamp;
+			cachedETag = metadata.etag;
+			cachedTag = metadata.tag;
+		}

-			if (age < CACHE_TTL && existsSync(CACHE_FILE)) {
+		// If tag changed, we need to fetch new instructions
+		if (cachedTag !== latestTag) {
+			cachedETag = null; // Force re-fetch
+		}
+
+		// Make conditional request with If-None-Match header
+		const headers = {};
+		if (cachedETag) {
+			headers["If-None-Match"] = cachedETag;
+		}
+
+		const response = await fetch(CODEX_INSTRUCTIONS_URL, { headers });
+
+		// 304 Not Modified - our cached version is still current
+		if (response.status === 304) {
+			if (existsSync(CACHE_FILE)) {
 				return readFileSync(CACHE_FILE, "utf8");
 			}
+			// Cache file missing but GitHub says not modified - fall through to re-fetch
 		}

-		// Fetch fresh instructions from GitHub
-		const response = await fetch(CODEX_INSTRUCTIONS_URL);
-		if (!response.ok) throw new Error(`HTTP ${response.status}`);
+		// 200 OK - new content or first fetch
+		if (response.ok) {
+			const instructions = await response.text();
+			const newETag = response.headers.get("etag");

-		const instructions = await response.text();
+			// Create cache directory if it doesn't exist
+			if (!existsSync(CACHE_DIR)) {
+				mkdirSync(CACHE_DIR, { recursive: true });
+			}

-		// Create cache directory if it doesn't exist
-		if (!existsSync(CACHE_DIR)) {
-			mkdirSync(CACHE_DIR, { recursive: true });
+			// Cache the instructions with ETag and tag (verbatim from GitHub)
+			writeFileSync(CACHE_FILE, instructions, "utf8");
+			writeFileSync(
+				CACHE_METADATA_FILE,
+				JSON.stringify({
+					etag: newETag,
+					tag: latestTag,
+					lastChecked: Date.now(),
+					url: CODEX_INSTRUCTIONS_URL,
+				}),
+				"utf8",
+			);
+
+			return instructions;
 		}

-		// Cache the instructions
-		writeFileSync(CACHE_FILE, instructions, "utf8");
-		writeFileSync(
-			CACHE_METADATA_FILE,
-			JSON.stringify({ timestamp: Date.now(), url: CODEX_INSTRUCTIONS_URL }),
-			"utf8",
-		);
-
-		return instructions;
+		throw new Error(`HTTP ${response.status}`);
 	} catch (error) {
 		console.error(
 			"[openai-codex-plugin] Failed to fetch instructions from GitHub:",
 			error.message,
 		);

-		// Try to use cached version even if expired
+		// Try to use cached version even if stale
 		if (existsSync(CACHE_FILE)) {
-			console.error("[openai-codex-plugin] Using stale cached instructions");
+			console.error("[openai-codex-plugin] Using cached instructions");
 			return readFileSync(CACHE_FILE, "utf8");
 		}

@@ -54,7 +54,7 @@ export function startLocalOAuthServer({ state }) {
 			})
 			.on("error", (err) => {
 				console.error(
-					"[openai-codex-auth] Failed to bind http://127.0.0.1:1455 (",
+					"[openai-codex-plugin] Failed to bind http://127.0.0.1:1455 (",
 					err?.code,
 					") Falling back to manual paste.",
 				);
@@ -1,6 +1,6 @@
 {
  "name": "opencode-openai-codex-auth",
-  "version": "1.0.1",
+  "version": "1.0.2",
  "description": "OpenAI ChatGPT (Codex backend) OAuth auth plugin for opencode - use your ChatGPT Plus/Pro subscription instead of API credits",
  "main": "./index.mjs",
  "type": "module",