Remove plugin-side compaction

README.md

@@ -52,9 +52,6 @@ Set these in `~/.opencode/openhax-codex-config.json` (applies to all models):
 
 - `codexMode` (default `true`): enable the Codex ↔ OpenCode bridge prompt and tool remapping
 - `enablePromptCaching` (default `true`): keep a stable `prompt_cache_key` so Codex can reuse cached prompts
-- `enableCodexCompaction` (default `true`): allow `/codex-compact` behavior once upstream support lands
-- `autoCompactTokenLimit` (optional): trigger Codex compaction after an approximate token threshold
-- `autoCompactMinMessages` (default `8`): minimum conversation turns before auto-compaction is considered
 
 Example:
 
@@ -565,8 +562,6 @@ If you want to customize settings yourself, you can configure options at provide
 >
 > † **Extra High reasoning**: `reasoningEffort: "xhigh"` provides maximum computational effort for complex, multi-step problems and is only available on `gpt-5.1-codex-max`.
 
-See [Plugin-Level Settings](#plugin-level-settings) above for global toggles. Below are provider/model examples.
-
 #### Global Configuration Example
 
 Apply settings to all models:

docs/configuration.md

@@ -47,21 +47,25 @@ Complete reference for configuring the OpenHax Codex Plugin.
 Controls computational effort for reasoning.
 
 **GPT-5 Values:**
+
 - `minimal` - Fastest, least reasoning
 - `low` - Light reasoning
 - `medium` - Balanced (default)
 - `high` - Deep reasoning
 
 **GPT-5-Codex Values:**
+
 - `low` - Fastest for code
 - `medium` - Balanced (default)
 - `high` - Maximum code quality
 
 **Notes**:
+
 - `minimal` auto-converts to `low` for gpt-5-codex (API limitation)
 - `gpt-5-codex-mini*` only supports `medium` or `high`; lower settings are clamped to `medium`
 
 **Example:**
+
 ```json
 {
   "options": {
@@ -75,10 +79,12 @@ Controls computational effort for reasoning.
 Controls reasoning summary verbosity.
 
 **Values:**
+
 - `auto` - Automatically adapts (default)
 - `detailed` - Verbose summaries
 
 **Example:**
+
 ```json
 {
   "options": {
@@ -92,14 +98,17 @@ Controls reasoning summary verbosity.
 Controls output length.
 
 **GPT-5 Values:**
+
 - `low` - Concise
 - `medium` - Balanced (default)
 - `high` - Verbose
 
 **GPT-5-Codex:**
+
 - `medium` only (API limitation)
 
 **Example:**
+
 ```json
 {
   "options": {
@@ -117,6 +126,7 @@ Array of additional response fields to include.
 **Why needed**: Enables multi-turn conversations with `store: false` (stateless mode)
 
 **Example:**
+
 ```json
 {
   "options": {
@@ -132,13 +142,15 @@ Controls server-side conversation persistence.
 **⚠️ Required**: `false` (for AI SDK 2.0.50+ compatibility)
 
 **Values:**
+
 - `false` - Stateless mode (required for Codex API)
 - `true` - Server-side storage (not supported by Codex API)
 
 **Why required:**
 AI SDK 2.0.50+ automatically uses `item_reference` items when `store: true`. The Codex API requires stateless operation (`store: false`), where references cannot be resolved.
 
 **Example:**
+
 ```json
 {
   "options": {
@@ -241,6 +253,7 @@ Different settings for different models:
 - **`id` field**: DEPRECATED - not used by OpenAI provider
 
 **Example Usage:**
+
 ```bash
 # Use the config key in CLI
 opencode run "task" --model=openai/my-custom-id
@@ -323,6 +336,7 @@ Different agents use different models:
 Global config has defaults, project overrides for specific work:
 
 **~/.config/opencode/opencode.json** (global):
+
 ```json
 {
   "plugin": ["@openhax/codex"],
@@ -338,6 +352,7 @@ Global config has defaults, project overrides for specific work:
 ```
 
 **my-project/.opencode.json** (project):
+
 ```json
 {
   "provider": {
@@ -362,55 +377,39 @@ Advanced plugin settings in `~/.opencode/openhax-codex-config.json`:
 ```json
 {
   "codexMode": true,
-  "enableCodexCompaction": true,
-  "autoCompactTokenLimit": 12000,
-  "autoCompactMinMessages": 8
+  "enablePromptCaching": true
 }
 ```
 
 ### Log file management
 
 Control local request/rolling log growth:
+
 - `CODEX_LOG_MAX_BYTES` (default: 5_242_880) - rotate when the rolling log exceeds this many bytes.
 - `CODEX_LOG_MAX_FILES` (default: 5) - number of rotated log files to retain (plus the active log).
 - `CODEX_LOG_QUEUE_MAX` (default: 1000) - maximum buffered log entries before oldest entries are dropped.
 
 ### CODEX_MODE
 
 **What it does:**
+
 - `true` (default): Uses Codex-OpenCode bridge prompt (Task tool & MCP aware)
 - `false`: Uses legacy tool remap message
 - Bridge prompt content is synced with the latest Codex CLI release (ETag-cached)
 
 **When to disable:**
+
 - Compatibility issues with OpenCode updates
 - Testing different prompt styles
 - Debugging tool call issues
 
 **Override with environment variable:**
+
 ```bash
 CODEX_MODE=0 opencode run "task"  # Temporarily disable
 CODEX_MODE=1 opencode run "task"  # Temporarily enable
 ```
 
-### enableCodexCompaction
-
-Controls whether the plugin exposes Codex-style compaction commands.
-
-- `true` (default): `/codex-compact` is available and auto-compaction heuristics may run if enabled.
-- `false`: Compaction commands are ignored and OpenCode's own prompts pass through untouched.
-
-Disable only if you prefer OpenCode's host-side compaction or while debugging prompt differences.
-
-### autoCompactTokenLimit / autoCompactMinMessages
-
-Configures the optional auto-compaction heuristic.
-
-- `autoCompactTokenLimit`: Approximate token budget (based on character count ÷ 4). When unset, auto-compaction never triggers.
-- `autoCompactMinMessages`: Minimum number of conversation turns before auto-compaction is considered (default `8`).
-
-When the limit is reached, the plugin injects a Codex summary, stores it for future turns, and replies: “Auto compaction triggered… Review the summary then resend your last instruction.”
-
 ### Prompt caching
 
 - When OpenCode provides a `prompt_cache_key` (its session identifier), the plugin forwards it directly to Codex.
@@ -429,12 +428,14 @@ When the limit is reached, the plugin injects a Codex summary, stores it for fut
 ## Configuration Files
 
 **Provided Examples:**
+
 - [config/full-opencode.json](../config/full-opencode.json) - Complete with 11 variants (adds Codex Mini presets)
 - [config/minimal-opencode.json](../config/minimal-opencode.json) - Minimal setup
 
 > **Why choose the full config?** OpenCode's auto-compaction and usage widgets rely on the per-model `limit` metadata present only in `full-opencode.json`. Use the minimal config only if you don't need those UI features.
 
 **Your Configs:**
+
 - `~/.config/opencode/opencode.json` - Global config
 - `<project>/.opencode.json` - Project-specific config
 - `~/.opencode/openhax-codex-config.json` - Plugin config
@@ -458,6 +459,7 @@ DEBUG_CODEX_PLUGIN=1 opencode run "test" --model=openai/your-model-name
 ```
 
 Look for:
+
 ```
 [openhax/codex] Model config lookup: "your-model-name" → normalized to "gpt-5-codex" for API {
   hasModelSpecificConfig: true,
@@ -512,6 +514,7 @@ Old verbose names still work:
 ```
 
 **Benefits:**
+
 - Cleaner: `--model=openai/gpt-5-codex-low`
 - Matches Codex CLI preset names
 - No redundant `id` field
@@ -608,9 +611,11 @@ Old verbose names still work:
 **Cause**: Config key doesn't match model name in command
 
 **Fix**: Use exact config key:
+
 ```json
 { "models": { "my-model": { ... } } }
 ```
+
 ```bash
 opencode run "test" --model=openai/my-model  # Must match exactly
 ```
@@ -630,9 +635,11 @@ Look for `hasModelSpecificConfig: true` in debug output.
 **Cause**: Model normalizes before lookup
 
 **Example Problem:**
+
 ```json
 { "models": { "gpt-5-codex": { "options": { ... } } } }
 ```
+
 ```bash
 --model=openai/gpt-5-codex-low  # Normalizes to "gpt-5-codex" before lookup
 ```

docs/getting-started.md

@@ -207,6 +207,7 @@ Add this to `~/.config/opencode/opencode.json`:
 ```
 
 **What you get:**
+
 - ✅ GPT-5 Codex (Low/Medium/High reasoning)
 - ✅ GPT-5 (Minimal/Low/Medium/High reasoning)
 - ✅ gpt-5-mini, gpt-5-nano (lightweight variants)
@@ -290,6 +291,7 @@ opencode
 ```
 
 **When to update:**
+
 - New features released
 - Bug fixes available
 - Security updates
@@ -313,6 +315,7 @@ For plugin development or testing unreleased changes:
 **Note**: Must point to `dist/` folder (built output), not root.
 
 **Build the plugin:**
+
 ```bash
 cd codex
 npm install
@@ -355,11 +358,13 @@ ls ~/.opencode/logs/codex-plugin/
 **Prompt caching is enabled by default** to minimize your costs.
 
 ### What This Means
+
 - Your conversation context is preserved across turns
 - Token usage is significantly reduced for multi-turn conversations
 - Lower overall costs compared to stateless operation
 
 ### Managing Caching
+
 Create `~/.opencode/openhax-codex-config.json`:
 
 ```json
@@ -369,27 +374,12 @@ Create `~/.opencode/openhax-codex-config.json`:
 ```
 
 **Settings:**
+
 - `true` (default): Optimize for cost savings
 - `false`: Fresh context each turn (higher costs)
 
 **⚠️ Warning**: Disabling caching will dramatically increase token usage and costs.
 
-### Compaction Controls
-
-To mirror the Codex CLI `/compact` command, add the following to `~/.opencode/openhax-codex-config.json`:
-
-```json
-{
-  "enableCodexCompaction": true,
-  "autoCompactTokenLimit": 12000,
-  "autoCompactMinMessages": 8
-}
-```
-
-- `enableCodexCompaction` toggles both the `/codex-compact` manual command and Codex-side history rewrites.
-- Set `autoCompactTokenLimit` to have the plugin run compaction automatically once the conversation grows beyond the specified budget.
-- Users receive the Codex summary (with the standard `SUMMARY_PREFIX`) and can immediately resend their paused instruction; subsequent turns are rebuilt from the stored summary instead of the entire backlog.
-
 ---
 
 ## Next Steps

lib/config.ts

@@ -12,8 +12,6 @@ const CONFIG_PATH = getOpenCodePath("openhax-codex-config.json");
 const DEFAULT_CONFIG: PluginConfig = {
 	codexMode: true,
 	enablePromptCaching: true,
-	enableCodexCompaction: true,
-	autoCompactMinMessages: 8,
 	logging: {
 		showWarningToasts: false,
 	},

lib/request/codex-fetcher.ts

@@ -1,7 +1,6 @@
 import type { PluginInput } from "@opencode-ai/plugin";
 import type { Auth } from "@opencode-ai/sdk";
 import { maybeHandleCodexCommand } from "../commands/codex-metrics.js";
-import { finalizeCompactionResponse } from "../compaction/compaction-executor.js";
 import { LOG_STAGES } from "../constants.js";
 import { logRequest } from "../logger.js";
 import { recordSessionResponseFromHandledResponse } from "../session/response-recorder.js";
@@ -93,16 +92,7 @@ export function createCodexFetcher(deps: CodexFetcherDeps) {
 			return await handleErrorResponse(response);
 		}
 
-		let handledResponse = await handleSuccessResponse(response, hasTools);
-
-		if (transformation?.compactionDecision) {
-			handledResponse = await finalizeCompactionResponse({
-				response: handledResponse,
-				decision: transformation.compactionDecision,
-				sessionManager,
-				sessionContext,
-			});
-		}
+		const handledResponse = await handleSuccessResponse(response, hasTools);
 
 		await recordSessionResponseFromHandledResponse({
 			sessionManager,