CHANGELOG.md

Changelog

All notable changes to this project are documented here. Dates use the ISO format (YYYY-MM-DD).

[4.4.0] - 2026-01-09

Maintenance release: OAuth success page version sync.

Changed

• OAuth success banner: Updates the success page header to display the current release version.

[4.3.1] - 2026-01-08

Installer safety release: JSONC support, safe uninstall, and minimal reasoning clamp.

Added

• JSONC-aware installer: preserves comments/formatting and prioritizes opencode.jsonc over opencode.json.
• Safe uninstall: --uninstall removes only plugin entries + our model presets; --all removes tokens/logs/cache.
• Installer tests: coverage for JSONC parsing, precedence, uninstall safety, and artifact cleanup.

Changed

• Default config path: installer creates ~/.config/opencode/opencode.jsonc when no config exists.
• Dependency: jsonc-parser added to keep JSONC updates robust and comment-safe.

Fixed

• Minimal reasoning clamp: minimal is now normalized to low for GPT‑5.x requests to avoid backend rejection.

[4.3.0] - 2026-01-04

Feature + reliability release: variants support, one-command installer, and auth/error handling fixes.

Added

• One-command installer/update: npx -y opencode-openai-codex-auth@latest (global config, backup, cache clear) with --legacy for OpenCode v1.0.209 and below.
• Modern variants config: config/opencode-modern.json for OpenCode v1.0.210+; legacy presets remain in config/opencode-legacy.json.
• Installer CLI bundled as package bin for cross-platform use (Windows/macOS/Linux).

Changed

• Variants-aware request config: respects host-supplied body.reasoning / providerOptions.openai before falling back to defaults.
• OpenCode prompt source: updates to the current upstream repository (anomalyco/opencode).
• Docs/README: install-first layout with leaner guidance and explicit legacy path.

Fixed

• Headless login fallback: missing xdg-open no longer fails the OAuth flow; manual URL paste stays available.
• Error handling alignment: refresh failures throw; usage-limit 404s map to retryable 429s where appropriate.
• AGENTS.md preservation: protected instruction markers stop accidental filtering of user instructions.
• Tool-call integrity: orphan outputs now match local_shell_call and custom_tool_call (Codex CLI parity); unmatched outputs preserved as assistant messages.
• Logging noise: debug logging gated behind flags to prevent stdout bleed.

[4.2.0] - 2025-12-19

Feature release: GPT 5.2 Codex support and prompt alignment with latest Codex CLI.

Added

• GPT 5.2 Codex model family: Full support for gpt-5.2-codex with presets:
  • gpt-5.2-codex-low - Fast GPT 5.2 Codex responses
  • gpt-5.2-codex-medium - Balanced GPT 5.2 Codex tasks
  • gpt-5.2-codex-high - Complex GPT 5.2 Codex reasoning & tools
  • gpt-5.2-codex-xhigh - Deep GPT 5.2 Codex long-horizon work
• New model family prompt: gpt-5.2-codex_prompt.md fetched from the latest Codex CLI release with its own cache file.
• Test coverage: Added unit tests for GPT 5.2 Codex normalization, family selection, and reasoning behavior.

Changed

• Prompt selection alignment: GPT 5.2 general now uses gpt_5_2_prompt.md (Codex CLI parity).
• Reasoning configuration: GPT 5.2 Codex supports xhigh but does not support "none"; "none" auto-upgrades to "low" and "minimal" normalizes to "low".
• Config presets: config/opencode-legacy.json includes the 22 pre-configured presets (adds GPT 5.2 Codex); config/opencode-modern.json provides the variant-based setup.
• Docs: Updated README/AGENTS/config docs to include GPT 5.2 Codex and new model family behavior.

[4.1.1] - 2025-12-17

Minor release: "none" reasoning effort support, orphaned function_call_output fix, and HTML version update.

Added

• "none" reasoning effort support: GPT-5.1 and GPT-5.2 support reasoning_effort: "none" which disables the reasoning phase entirely. This can result in faster responses when reasoning is not needed.
  • gpt-5.2-none - GPT-5.2 with reasoning disabled
  • gpt-5.1-none - GPT-5.1 with reasoning disabled
• 4 new unit tests for "none" reasoning behavior (now 197 total unit tests).

Fixed

• Orphaned function_call_output 400 errors: Fixed API errors when conversation history contains item_reference pointing to stored function calls. Previously, orphaned function_call_output items were only filtered when !body.tools. Now always handles orphans regardless of tools presence, and converts them to assistant messages to preserve context while avoiding API errors.
• OAuth HTML version display: Updated version in oauth-success.html from 1.0.4 to 4.1.0.

Technical Details

• getReasoningConfig() now detects GPT-5.1 general purpose models (not Codex variants) and allows "none" to pass through.
• GPT-5.2 inherits "none" support as it's newer than GPT-5.1.
• Codex variants (gpt-5.1-codex, gpt-5.1-codex-max, gpt-5.1-codex-mini) do NOT support "none":
  • Codex and Codex Max: "none" auto-converts to "low"
  • Codex Mini: "none" auto-converts to "medium" (as before)
• Documentation updated with complete reasoning effort support matrix per model family.

References

• OpenAI API docs (platform.openai.com/docs/api-reference/chat/create): "gpt-5.1 defaults to none, which does not perform reasoning. The supported reasoning values for gpt-5.1 are none, low, medium, and high."
• Codex CLI (codex-rs/protocol/src/openai_models.rs): ReasoningEffort enum includes None variant with #[serde(rename_all = "lowercase")] serialization to "none".
• Codex CLI (codex-rs/core/src/client.rs): Request builder passes ReasoningEffort::None through to API without validation/rejection.
• Codex CLI (docs/config.md): Documents model_reasoning_effort = "none" as valid config option.

Notes

• This plugin defaults to "medium" for better coding assistance; users must explicitly set "none" if desired.

[4.1.0] - 2025-12-11

Feature release: GPT 5.2 model support and image input capabilities.

Added

• GPT 5.2 model family support: Full support for OpenAI's latest GPT 5.2 model with 4 reasoning level presets:
  • gpt-5.2-low - Fast responses with light reasoning
  • gpt-5.2-medium - Balanced reasoning for general tasks
  • gpt-5.2-high - Complex reasoning and analysis
  • gpt-5.2-xhigh - Deep multi-hour analysis (same as Codex Max)
• Full image input support: All 16 model variants now include modalities.input: ["text", "image"] enabling full multimodal capabilities - read screenshots, diagrams, UI mockups, and any image directly in OpenCode.
• GPT 5.2 model family added to codex.ts with dedicated prompt handling.
• Test coverage: Updated integration tests to verify all 16 models (was 13), now 193 unit tests + 16 integration tests.

Changed

• Model ordering: Config now ordered by model family priority: GPT 5.2 → Codex Max → Codex → Codex Mini → GPT 5.1.
• Removed default presets: Removed gpt-5.1-codex-max and gpt-5.2 (without reasoning suffix) to enforce explicit reasoning level selection.
• Test script: scripts/test-all-models.sh now uses local dist for testing and includes GPT 5.2 tests.
• Documentation: Updated README with GPT 5.2 models, image support, and condensed config example.

Technical Details

• GPT 5.2 maps to gpt-5.2 API model with same reasoning options as Codex Max (low/medium/high/xhigh).
• getModelFamily() now returns "gpt-5.2" for GPT 5.2 models, using Codex Max prompts.
• getReasoningConfig() treats GPT 5.2 like Codex Max for xhigh reasoning support.
• Model normalization pattern matching updated to recognize GPT 5.2 before other patterns.

[4.0.2] - 2025-11-27

Bugfix release: Fixes compaction context loss, agent creation, and SSE/JSON response handling.

Fixed

• Compaction losing context: v4.0.1 was too aggressive in filtering tool calls - it removed ALL function_call/function_call_output items when tools weren't present. Now only orphaned outputs (without matching calls) are filtered, preserving matched pairs for compaction context.
• Agent creation failing: The /agent create command was failing with "Invalid JSON response" because we were returning SSE streams instead of JSON for generateText() requests.
• SSE/JSON response handling: Properly detect original request intent - streamText() requests get SSE passthrough, generateText() requests get SSE→JSON conversion.

Added

• gpt-5.1-chat-latest model support: Added to model map, normalizes to gpt-5.1.

Technical Details

• Root cause of compaction issue: OpenCode sends item_reference with fc_* IDs for function calls. We filter these for stateless mode, but v4.0.1 then removed ALL tool items. Now we only remove orphaned function_call_output items (where no matching function_call exists).
• Root cause of agent creation issue: We were forcing stream: true for all requests and returning SSE for all responses. Now we capture original stream value before transformation and convert SSE→JSON only when original request wasn't streaming.
• The Codex API always receives stream: true (required), but response handling is based on original intent.

[4.0.1] - 2025-11-27

Bugfix release: Fixes API errors during summary/compaction and GitHub rate limiting.

Fixed

• Orphaned function_call_output errors: Fixed 400 errors during summary/compaction requests when OpenCode sends item_reference pointers to server-stored function calls. The plugin now filters out function_call and function_call_output items when no tools are present in the request.
• GitHub API rate limiting: Added fallback mechanism when fetching Codex instructions from GitHub. If the API returns 403 (rate limit), the plugin now falls back to parsing the HTML releases page.

Technical Details

• Root cause: OpenCode's secondary model (gpt-5-nano) uses item_reference with fc_* IDs to reference stored function calls. Our plugin filters item_reference for stateless mode (store: false), leaving function_call_output orphaned. The Codex API rejects requests with orphaned outputs.
• Fix: When hasTools === false, filter out all function_call and function_call_output items from the input array.
• GitHub fallback chain: API endpoint → HTML page → redirect URL parsing → HTML regex parsing.

[4.0.0] - 2025-11-25

Major release: Complete prompt engineering overhaul matching official Codex CLI behavior, with full GPT-5.1 Codex Max support.

Highlights

• Full Codex Max support with dedicated prompt including frontend design guidelines
• Model-specific prompts matching Codex CLI's prompt selection logic
• GPT-5.0 → GPT-5.1 migration as legacy models are phased out

Added

• Model-specific system prompts: Plugin now fetches the correct Codex prompt based on model family, matching Codex CLI's model_family.rs logic:
  • gpt-5.1-codex-max* → gpt-5.1-codex-max_prompt.md (117 lines, includes frontend design guidelines)
  • gpt-5.1-codex*, gpt-5.1-codex-mini* → gpt_5_codex_prompt.md (105 lines, focused coding prompt)
  • gpt-5.1* → gpt_5_1_prompt.md (368 lines, full behavioral guidance)
• New ModelFamily type ("codex-max" | "codex" | "gpt-5.1") for prompt selection.
• New getModelFamily() function to determine prompt selection based on normalized model name.
• Model family now logged in request logs for debugging (modelFamily field in after-transform logs).
• 16 new unit tests for model family detection (now 191 total unit tests).
• Integration tests now verify correct model family selection (13 integration tests with family verification).

Changed

• Legacy GPT-5.0 models now map to GPT-5.1: All legacy gpt-5 model variants automatically normalize to their gpt-5.1 equivalents as GPT-5.0 is being phased out by OpenAI:
  • gpt-5-codex → gpt-5.1-codex
  • gpt-5 → gpt-5.1
  • gpt-5-mini, gpt-5-nano → gpt-5.1
  • codex-mini-latest → gpt-5.1-codex-mini
• Lazy instruction loading: Instructions are now fetched per-request based on model family (not pre-loaded at initialization).
• Separate caching per model family: Each model family has its own cached prompt file:
  • codex-max-instructions.md + codex-max-instructions-meta.json
  • codex-instructions.md + codex-instructions-meta.json
  • gpt-5.1-instructions.md + gpt-5.1-instructions-meta.json

Fixed

• Fixed OpenCode prompt cache URL to fetch from dev branch instead of non-existent main branch.
• Fixed model configuration test script to correctly identify model logs in multi-model sessions (opencode uses a small model like gpt-5-nano for title generation alongside the user's selected model).

Technical Details

This release brings full parity with Codex CLI's prompt engineering:

• Codex family (105 lines): Concise, tool-focused prompt for coding tasks
• Codex Max family (117 lines): Adds frontend design guidelines for UI work
• GPT-5.1 general (368 lines): Comprehensive behavioral guidance, personality, planning

[3.3.0] - 2025-11-19

Added

• GPT 5.1 Codex Max support: normalization, per-model defaults, and new presets (gpt-5.1-codex-max, gpt-5.1-codex-max-xhigh) with extended reasoning options (including none/xhigh) while keeping the 272k context / 128k output limits.
• Typing and config support for new reasoning options (none/xhigh, summary off/on) plus updated test matrix entries.

Changed

• Codex Mini clamping now downgrades unsupported xhigh to high and guards against none/minimal inputs.
• Documentation, config guides, and validation scripts now reflect 13 verified GPT 5.1 variants (3 codex, 5 codex-max, 2 codex-mini, 3 general), including Codex Max. See README for details on pre-configured variants.

[3.2.0] - 2025-11-14

Added

• GPT 5.1 model family support: normalization for gpt-5.1, gpt-5.1-codex, and gpt-5.1-codex-mini plus new GPT 5.1-only presets in the canonical config/opencode-legacy.json.
• Documentation updates (README, docs, AGENTS) describing the 5.1 families, their reasoning defaults, and how they map to ChatGPT slugs and token limits.

Changed

• Model normalization docs and tests now explicitly cover both 5.0 and 5.1 Codex/general families and the two Codex Mini tiers.
• The legacy GPT 5.0 full configuration is now published separately; new installs should prefer the 5.1 presets in config/opencode-legacy.json.

[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/opencode-legacy.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.