Merge pull request #10 from ndycode/fix/multi-account-duplicate-accountid

feat: v4.6.0 - Context overflow handling and missing tool result injection

Author: ndycode

CHANGELOG.md

@@ -8,15 +8,27 @@ All notable changes to this project are documented here. Dates use the ISO forma
 
 ### Changed
 
-## [4.3.1] - 2026-01-23
+## [4.6.0] - 2026-01-25
+
+**Feature release**: Context overflow handling and missing tool result injection.
 
 ### Added
-- `openai-accounts-status --json` for scriptable status output with email/ID labels.
+- **Context Overflow Handler**: Gracefully handles "prompt too long" / context length exceeded errors:
+  - Returns synthetic SSE response with helpful instructions instead of raw 400 error
+  - Suggests `/compact`, `/clear`, or `/undo` commands to reduce context size
+  - Prevents OpenCode session from getting locked on context overflow
+  - New module: `lib/context-overflow.ts`
+- **Missing Tool Result Injection**: Automatically handles cancelled tool calls (ESC mid-execution):
+  - Detects orphaned `function_call` items (calls without matching outputs)
+  - Injects synthetic output: `"Operation cancelled by user"`
+  - Prevents "missing tool_result" API errors when user cancels mid-tool
+  - New function: `injectMissingToolOutputs()` in `lib/request/helpers/input-utils.ts`
+- **34 new unit tests** for context overflow and tool injection (now 379 total tests)
 
-### Changed
-- Account labels now prefer email and show ID suffix when available; list/status outputs are columnized for readability.
-- Stored account emails are trimmed/lowercased when present.
-- Dependency refresh: @opencode-ai plugin/sdk 1.1.34, hono 4.11.5, vitest 4.0.18, @types/node 25.0.10, @typescript-eslint 8.53.1.
+### Technical Details
+- Context overflow detection matches patterns: `prompt_too_long`, `context_length_exceeded`, `maximum context length`, `token limit exceeded`, `too many tokens`
+- Synthetic SSE response includes proper message_start/content_block_delta/message_stop events
+- Tool injection preserves message order: outputs are placed immediately after their calls
 
 ## [4.5.0] - 2026-01-25
 
@@ -48,6 +60,16 @@ All notable changes to this project are documented here. Dates use the ISO forma
 - Stale entries (>30s) are automatically cleaned up to prevent memory leaks
 - Auto-update check is non-blocking and fails silently to avoid disrupting plugin operation
 
+## [4.3.1] - 2026-01-23
+
+### Added
+- `openai-accounts-status --json` for scriptable status output with email/ID labels.
+
+### Changed
+- Account labels now prefer email and show ID suffix when available; list/status outputs are columnized for readability.
+- Stored account emails are trimmed/lowercased when present.
+- Dependency refresh: @opencode-ai plugin/sdk 1.1.34, hono 4.11.5, vitest 4.0.18, @types/node 25.0.10, @typescript-eslint 8.53.1.
+
 ## [4.4.0] - 2026-01-25
 
 **Feature release**: Intelligent rate-limit rotation with health-based account selection.

README.md

@@ -54,7 +54,7 @@ Install the opencode-openai-codex-auth-multi plugin and add the OpenAI model def
 **Option B: One-command install**
 
 ```bash
-npx -y opencode-openai-codex-auth-multi@4.3.1
+npx -y opencode-openai-codex-auth-multi@latest
 ```
 
 This writes the config to `~/.config/opencode/opencode.json`, backs up existing config, and clears the plugin cache.
@@ -67,7 +67,7 @@ This writes the config to `~/.config/opencode/opencode.json`, backs up existing
 
    ```json
    {
-     "plugin": ["opencode-openai-codex-auth-multi@4.3.1"]
+     "plugin": ["opencode-openai-codex-auth-multi@latest"]
    }
    ```
 
@@ -99,7 +99,7 @@ This writes the config to `~/.config/opencode/opencode.json`, backs up existing
 2. Add the plugin to the `plugin` array:
    ```json
    {
-     "plugin": ["opencode-openai-codex-auth-multi@4.3.1"]
+     "plugin": ["opencode-openai-codex-auth-multi@latest"]
    }
    ```
 
@@ -147,7 +147,7 @@ Add this to your `~/.config/opencode/opencode.json`:
 ```json
 {
   "$schema": "https://opencode.ai/config.json",
-  "plugin": ["opencode-openai-codex-auth-multi@4.3.1"],
+  "plugin": ["opencode-openai-codex-auth-multi@latest"],
   "provider": {
     "openai": {
       "options": {
@@ -236,7 +236,7 @@ For legacy OpenCode (v1.0.209 and below), use `config/opencode-legacy.json` whic
 
 ## Multi-Account Setup
 
-Add multiple ChatGPT accounts for higher combined quotas. The plugin automatically rotates between accounts when one is rate-limited.
+Add multiple ChatGPT accounts for higher combined quotas. The plugin uses **health-aware rotation** with automatic failover.
 
 ```bash
 opencode auth login  # Run again to add more accounts
@@ -247,6 +247,12 @@ opencode auth login  # Run again to add more accounts
 - `openai-accounts-switch` — Switch active account
 - `openai-accounts-status` — Show rate limit status
 
+**How rotation works (v4.4.0+):**
+- Health scoring tracks success/failure per account
+- Token bucket prevents hitting rate limits
+- Hybrid selection prefers healthy accounts with available tokens
+- Always retries when all accounts are rate-limited (waits for reset)
+
 **Storage:** `~/.opencode/openai-codex-accounts.json`
 
 ---
@@ -354,7 +360,7 @@ OpenCode uses `~/.config/opencode/` on **all platforms** including Windows.
 **Solutions:**
 1. Update plugin:
    ```bash
-   npx -y opencode-openai-codex-auth-multi@4.3.1
+npx -y opencode-openai-codex-auth-multi@latest
    ```
 2. Ensure config has:
    ```json
@@ -399,7 +405,7 @@ Works alongside oh-my-opencode. No special configuration needed.
 ```json
 {
   "plugin": [
-    "opencode-openai-codex-auth-multi@4.3.1",
+    "opencode-openai-codex-auth-multi@latest",
     "oh-my-opencode@latest"
   ]
 }
@@ -412,7 +418,7 @@ List this plugin BEFORE DCP:
 ```json
 {
   "plugin": [
-    "opencode-openai-codex-auth-multi@4.3.1",
+    "opencode-openai-codex-auth-multi@latest",
     "@tarquinen/opencode-dcp@latest"
   ]
 }
@@ -434,6 +440,14 @@ Create `~/.opencode/openai-codex-auth-config.json` for optional settings:
 |--------|---------|--------------|
 | `codexMode` | `true` | Uses Codex-OpenCode bridge prompt (synced with latest Codex CLI) |
 
+### Retry Behavior (v4.4.0+)
+
+| Option | Default | What it does |
+|--------|---------|--------------|
+| `retryAllAccountsRateLimited` | `true` | Wait and retry when all accounts are rate-limited |
+| `retryAllAccountsMaxWaitMs` | `0` | Max wait time (0 = unlimited) |
+| `retryAllAccountsMaxRetries` | `Infinity` | Max retry attempts |
+
 ### Environment Variables
 
 ```bash

docs/configuration.md

@@ -220,7 +220,10 @@ Advanced plugin settings in `~/.opencode/openai-codex-auth-config.json`:
 
 ```json
 {
-  "codexMode": true
+  "codexMode": true,
+  "retryAllAccountsRateLimited": true,
+  "retryAllAccountsMaxWaitMs": 0,
+  "retryAllAccountsMaxRetries": null
 }
 ```
 
@@ -229,15 +232,22 @@ Advanced plugin settings in `~/.opencode/openai-codex-auth-config.json`:
 | Option | Default | What it does |
 |--------|---------|--------------|
 | `codexMode` | `true` | Uses Codex-OpenCode bridge prompt (synced with Codex CLI) |
+| `retryAllAccountsRateLimited` | `true` | Wait and retry when all accounts are rate-limited |
+| `retryAllAccountsMaxWaitMs` | `0` | Max wait time in ms (0 = unlimited) |
+| `retryAllAccountsMaxRetries` | `Infinity` | Max retry attempts (null = unlimited) |
 
 ### Environment Variables
 
 | Variable | Description |
 |----------|-------------|
 | `DEBUG_CODEX_PLUGIN=1` | Enable debug logging |
 | `ENABLE_PLUGIN_REQUEST_LOGGING=1` | Log all API requests |
+| `CODEX_PLUGIN_LOG_LEVEL=debug` | Set log level (debug, info, warn, error) |
 | `CODEX_MODE=0` | Temporarily disable bridge prompt |
 | `CODEX_MODE=1` | Temporarily enable bridge prompt |
+| `CODEX_AUTH_RETRY_ALL_RATE_LIMITED=0` | Disable wait-and-retry behavior |
+| `CODEX_AUTH_RETRY_ALL_MAX_WAIT_MS=30000` | Set max wait time |
+| `CODEX_AUTH_RETRY_ALL_MAX_RETRIES=1` | Set max retries |
 
 ---
 

docs/development/ARCHITECTURE.md

@@ -428,6 +428,62 @@ let include: Vec<String> = if reasoning.is_some() {
 
 ---
 
+## Multi-Account Rotation (v4.4.0+)
+
+### Health-Based Account Selection
+
+The plugin tracks account health and uses intelligent rotation:
+
+```
+Account Selection Flow:
+1. Score = (health × 2) + (tokens × 5) + (freshness × 0.1)
+2. Select account with highest score
+3. Consume token from bucket
+4. On success: health +1
+5. On rate limit: health -10, mark rate-limited
+6. On failure: health -20
+7. Passive recovery: +2 health/hour
+```
+
+### Token Bucket Rate Limiting
+
+Client-side rate limiting prevents hitting API limits:
+
+| Parameter | Value |
+|-----------|-------|
+| Max tokens | 50 |
+| Regeneration | 6 tokens/min |
+| Consume per request | 1 token |
+
+### Reason-Aware Backoff
+
+Different rate limit reasons use different backoff multipliers:
+
+| Reason | Multiplier | Description |
+|--------|------------|-------------|
+| `quota` | 3.0× | Daily quota exhausted |
+| `tokens` | 1.5× | Token limit hit |
+| `concurrent` | 0.5× | Concurrent request limit |
+| `unknown` | 1.0× | Default |
+
+### RefreshQueue (v4.5.0+)
+
+Prevents race conditions when multiple concurrent requests try to refresh the same token:
+
+```typescript
+// Without RefreshQueue: N concurrent requests = N refresh attempts
+// With RefreshQueue: N concurrent requests = 1 refresh, N-1 await
+
+const queue = getRefreshQueue();
+const tokens = await queue.queuedRefresh(refreshToken, async () => {
+  return await actualRefresh(refreshToken);
+});
+```
+
+**Source**: `lib/refresh-queue.ts`, `lib/rotation.ts`
+
+---
+
 ## Performance Considerations
 
 ### Token Usage

index.ts

@@ -55,6 +55,7 @@ import {
 } from "./lib/constants.js";
 import { logRequest, logDebug } from "./lib/logger.js";
 import { checkAndNotify } from "./lib/auto-update-checker.js";
+import { handleContextOverflow } from "./lib/context-overflow.js";
 import {
         AccountManager,
         extractAccountEmail,
@@ -660,10 +661,16 @@ export const OpenAIAuthPlugin: Plugin = async ({ client }: PluginInput) => {
 																											headers: Object.fromEntries(response.headers.entries()),
 																										});
 
-																											if (!response.ok) {
-																												const { response: errorResponse, rateLimit } =
-																													await handleErrorResponse(response);
-																												if (rateLimit) {
+								if (!response.ok) {
+									// Check for context overflow (400 "prompt too long") before other error handling
+									const contextOverflowResult = await handleContextOverflow(response, model);
+									if (contextOverflowResult.handled) {
+										return contextOverflowResult.response;
+									}
+
+									const { response: errorResponse, rateLimit } =
+										await handleErrorResponse(response);
+									if (rateLimit) {
 																														const { attempt, delayMs } = getRateLimitBackoff(
 																															account.index,
 																															quotaKey,