Remove compaction paths and opt-in warning logging

.gitignore

@@ -1,5 +1,6 @@
 node_modules/
 bun.lockb
+.worktrees/
 
 dist/
 coverage/
@@ -17,4 +18,3 @@ tmp
 .worktrees/
 .envrc
 .env
-

CHANGELOG.md

@@ -3,89 +3,124 @@
 All notable changes to this project are documented here. Dates use the ISO format (YYYY-MM-DD).
 
 ## [3.3.0] - 2025-11-19
+
 ### Added
+
 - Codex Max support that mirrors the Codex CLI: normalization for every `gpt-5.1-codex-max` alias, `reasoningEffort: "xhigh"`, and unit tests covering both the transformer and request body integration path.
 - Documentation and configuration updates calling out Codex Max as the flagship preset, plus refreshed samples showing how to opt into the Extra High reasoning mode.
 
 ### Changed
+
 - Sample configs (`full` + `minimal`), README tables, AGENTS.md, and the diagnostics script now prefer `gpt-5.1-codex-max`, keeping plugin defaults aligned with Codex CLI behaviour.
 
 ### Fixed
+
 - Requests that specify `reasoningEffort: "xhigh"` for non-supported models are now automatically downgraded to `high`, preventing API errors when Codex Max isn't selected.
 
 ## [3.2.0] - 2025-11-13
+
 ### Added
+
 - GPT-5.1 family integration: normalization for `gpt-5.1`/`gpt-5.1-codex`/`gpt-5.1-codex-mini`, expanded reasoning heuristics (including `reasoningEffort: "none"`), and preservation of the native `shell`/`apply_patch` tools emitted by Codex CLI.
 - Updated configuration, diagnostics script, and docs to showcase the 5.1 lineup (low/medium/high plus `none`) while keeping GPT-5 presets available for backwards compatibility.
 
 ### Changed
+
 - Default fallback model now targets `gpt-5.1`, and Codex Mini requests always use the new `gpt-5.1-codex-mini` slug to stay in sync with the latest Codex release.
 
 ### Fixed
+
 - Prevented invalid reasoning combinations by clamping unsupported `none`/`minimal` requests on Codex models and ensuring parallel tool-call behavior matches both GPT-5 and GPT-5.1 Codex variants.
 
 ## [3.1.0] - 2025-11-11
+
 ### Added
+
 - Codex Mini support end-to-end: normalization to the `codex-mini-latest` slug, proper reasoning defaults, and two new presets (`gpt-5-codex-mini-medium` / `gpt-5-codex-mini-high`).
 - Documentation & configuration updates describing the Codex Mini tier (200k input / 100k output tokens) plus refreshed totals (11 presets, 160+ unit tests).
 
 ### Fixed
+
 - Prevented Codex Mini from inheriting the lightweight (`minimal`) reasoning profile used by `gpt-5-mini`/`nano`, ensuring the API always receives supported effort levels.
 
 ## [3.0.0] - 2025-11-04
+
 ### Added
+
 - Codex-style usage-limit messaging that mirrors the 5-hour and weekly windows reported by the Codex CLI.
-- Documentation guidance noting that OpenCode's context auto-compaction and usage sidebar require the canonical `config/full-opencode.json`.
+- Documentation guidance noting that OpenCode's usage sidebar requires the canonical `config/full-opencode.json`.
 
 ### Changed
+
 - Prompt caching now relies solely on the host-supplied `prompt_cache_key`; conversation/session headers are forwarded only when OpenCode provides one.
 - CODEX_MODE bridge prompt refreshed to the newest Codex CLI release so tool awareness stays in sync.
 
 ### Fixed
+
 - Clarified README, docs, and configuration references so the canonical config matches shipped behaviour.
 - Pinned `hono` (4.10.4) and `vite` (7.1.12) to resolve upstream security advisories.
 
 ## [2.1.2] - 2025-10-12
+
 ### Added
+
 - Comprehensive compliance documentation (ToS guidance, security, privacy) and a full user/developer doc set.
 
 ### Fixed
+
 - Per-model configuration lookup, stateless multi-turn conversations, case-insensitive model normalization, and GitHub instruction caching.
 
 ## [2.1.1] - 2025-10-04
+
 ### Fixed
+
 - README cache-clearing snippet now runs in a subshell from the home directory to avoid path issues while removing cached plugin files.
 
 ## [2.1.0] - 2025-10-04
+
 ### Added
+
 - Enhanced CODEX_MODE bridge prompt with Task tool and MCP awareness plus ETag-backed verification of OpenCode system prompts.
 
 ### Changed
+
 - Request transformation made async to support prompt verification caching; AGENTS.md renamed to provide cross-agent guidance.
 
 ## [2.0.0] - 2025-10-03
+
 ### Added
+
 - Full TypeScript rewrite with strict typing, 123 automated tests, and nine pre-configured model variants matching the Codex CLI.
 - CODEX_MODE introduced (enabled by default) with a lightweight bridge prompt and configurability via config file or `CODEX_MODE` env var.
 
 ### Changed
+
 - Library reorganized into semantic folders (auth, prompts, request, etc.) and OAuth flow polished with the new success page.
 
 ## [1.0.3] - 2025-10-02
+
 ### Changed
+
 - Major internal refactor splitting the runtime into focused modules (logger, request/response handlers) and removing legacy debug output.
 
 ## [1.0.2] - 2025-10-02
+
 ### Added
+
 - ETag-based GitHub caching for Codex instructions and release-tag tracking for more stable prompt updates.
 
 ### Fixed
+
 - Default model fallback, text verbosity initialization, and standardized error logging prefixes.
 
 ## [1.0.1] - 2025-10-01
+
 ### Added
+
 - README clarifications: opencode auto-installs plugins, config locations, and streamlined quick-start instructions.
 
 ## [1.0.0] - 2025-10-01
+
 ### Added
+
 - Initial production release with ChatGPT Plus/Pro OAuth support, tool remapping, auto-updating Codex instructions, and zero runtime dependencies.

README.md

@@ -48,20 +48,33 @@ Want to customize? Jump to [Configuration reference](#configuration-reference).
 
 ## Plugin-Level Settings
 
-Set these in `~/.opencode/openhax-codex-config.json` (applies to all models):
+Set these in `~/.opencode/openhax-codex-config.json` (applies to all models). Related env vars control runtime tweaks (e.g., request logging, env tail):
 
 - `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
+- `logging` (optional): override log defaults and related env vars (`ENABLE_PLUGIN_REQUEST_LOGGING`, `DEBUG_CODEX_PLUGIN`, `CODEX_LOG_MAX_BYTES`, `CODEX_LOG_MAX_FILES`, `CODEX_LOG_QUEUE_MAX`, `CODEX_SHOW_WARNING_TOASTS`, `CODEX_LOG_WARNINGS_TO_CONSOLE`). Fields:
+  - `enableRequestLogging`: force request log persistence even without `ENABLE_PLUGIN_REQUEST_LOGGING=1`
+  - `debug`: force debug logging regardless of env
+  - `showWarningToasts`: show warning-level toasts in the OpenCode UI
+  - `logWarningsToConsole`: mirror warnings to console when toasts are off
+  - `logMaxBytes` (default `5_242_880` bytes): rotate rolling log after this size
+  - `logMaxFiles` (default `5`): rotated log files to retain (plus the active log)
+  - `logQueueMax` (default `1000`): max buffered log entries before oldest entries drop
+- Env tail (optional): set `CODEX_APPEND_ENV_CONTEXT=1` to reattach env/files context as a trailing developer message (stripped from system prompts to keep the prefix stable). Default is unset/0 (env/files removed for maximum cache stability).
+- Log inspection helper: `node scripts/inspect-codex-logs.mjs [--dir <path>] [--limit N] [--id X] [--stage after-transform]` summarizes cached request logs (shows model, prompt_cache_key, roles, etc.).
 
 Example:
 
 ```json
 {
   "codexMode": true,
   "enablePromptCaching": true,
-  "enableCodexCompaction": true,
-  "autoCompactTokenLimit": 120000,
-  "autoCompactMinMessages": 8
+  "logging": {
+    "enableRequestLogging": true,
+    "logMaxBytes": 5242880,
+    "logMaxFiles": 5,
+    "logQueueMax": 1000
+  }
 }
 ```
 
@@ -85,6 +98,8 @@ Example:
 
 **Prompt caching is enabled by default** to optimize your token usage and reduce costs.
 
+> Optional: `CODEX_APPEND_ENV_CONTEXT=1` keeps env/files context by reattaching it as a trailing developer message while preserving a stable prefix. Leave unset to maximize cache stability.
+
 ### How Caching Works
 
 - **Enabled by default**: `enablePromptCaching: true`
@@ -93,6 +108,14 @@ Example:
 - **Reduces token consumption** by reusing cached prompts
 - **Lowers costs** significantly for multi-turn conversations
 
+### Reducing Cache Churn (keep `prompt_cache_key` stable)
+
+- Why caches reset: OpenCode rebuilds the system/developer prompt every turn; the env block includes today’s date and a ripgrep tree of your workspace, so daily rollovers or file tree changes alter the prefix and trigger a new cache key.
+- Keep the tree stable: ensure noisy/ephemeral dirs are ignored (e.g. `dist/`, `build/`, `.next/`, `coverage/`, `.cache/`, `logs/`, `tmp/`, `.turbo/`, `.vite/`, `.stryker-tmp/`, `artifacts/`, and similar). Put transient outputs under an ignored directory or `/tmp`.
+- Don’t thrash the workspace mid-session: large checkouts, mass file generation, or moving directories will change the ripgrep listing and force a cache miss.
+- Model/provider switches also change the system prompt (different base prompt), so avoid swapping models in the middle of a session if you want to reuse cache.
+- Optional: set `CODEX_APPEND_ENV_CONTEXT=1` to reattach env/files at the end of the prompt instead of stripping them. This keeps the shared prefix stable (better cache reuse) while still sending env/files as a trailing developer message. Default is off (env/files stripped to maximize stability).
+
 ### Managing Caching
 
 #### Recommended: Full Configuration (Codex CLI Experience)

docs/code-cleanup-summary.md

@@ -13,7 +13,7 @@
 2. **Created InputItemUtils** - `lib/utils/input-item-utils.ts`
    - Centralized text extraction logic used in multiple modules
    - Added utility functions for role checking, filtering, and formatting
-   - Eliminates duplication in `request-transformer.ts`, `session-manager.ts`, and `codex-compaction.ts`
+   - Eliminates duplication in `request-transformer.ts` and `session-manager.ts`
    - Functions: `extractTextFromItem()`, `hasTextContent()`, `formatRole()`, `formatEntry()`, `isSystemMessage()`, `isUserMessage()`, `isAssistantMessage()`, `filterByRole()`, `getLastUserMessage()`, `countConversationTurns()`
 
 3. **Refactored Large Functions**
@@ -53,12 +53,14 @@
 ## Code Quality Improvements
 
 ### Before Refactoring
+
 - **Code Duplication**: 3+ duplicate clone implementations
 - **Large Functions**: `transformRequestBody()` 1130 lines with high complexity
 - **Magic Numbers**: Scattered TTL values and limits throughout codebase
 - **No Complexity Enforcement**: No cognitive complexity limits
 
 ### After Refactoring
+
 - **Eliminated Duplication**: Single source of truth for cloning and text extraction
 - **Reduced Complexity**: Large function now uses focused utility functions
 - **Centralized Configuration**: All magic numbers in constants with descriptive names
@@ -67,44 +69,50 @@
 ## Files Modified
 
 ### New Files Created
+
 - `lib/utils/clone.ts` - Shared cloning utilities
 - `lib/utils/input-item-utils.ts` - InputItem processing utilities
 
 ### Files Updated
+
 - `lib/constants.ts` - Added centralized configuration constants
 - `biome.json` - Enhanced linting rules for complexity
 - `lib/request/request-transformer.ts` - Updated to use shared utilities
 - `lib/session/session-manager.ts` - Updated to use shared utilities and constants
-- `lib/compaction/codex-compaction.ts` - Updated to use shared utilities
 - `test/session-manager.test.ts` - Updated imports for new constants
 
 ## Impact
 
 ### Maintainability
+
 - **Easier to modify** cloning behavior in one place
 - **Clearer separation of concerns** with focused utility functions
 - **Better discoverability** of common operations
 
 ### Performance
+
 - **Optimized cloning** with `structuredClone` when available
 - **Reduced memory allocation** through shared utilities
 - **Consistent error handling** patterns
 
 ### Code Quality
+
 - **Enforced complexity limits** to prevent future issues
 - **Standardized patterns** across all modules
 - **Improved type safety** with centralized utilities
 
 ## Next Steps
 
 The codebase now has:
+
 - **B+ code quality rating** (improved from existing baseline)
 - **Zero critical code smells**
 - **Comprehensive test coverage** maintained
 - **Automated quality gates** in place
 
 Future development will benefit from:
+
 - Shared utilities reducing duplication
 - Complexity limits preventing excessive nesting
 - Centralized configuration for easy maintenance
-- Consistent patterns across all modules
+- Consistent patterns across all modules