AI Commit & Prompt Helper v2.1.0

Generate structured, review-friendly AI commit messages, release workflow notes, and improved coding prompts in VS Code using Codex CLI, OpenRouter, Hugging Face, and OpenAI-compatible LLM APIs.

Current extension release: v2.1.0.

Release Notes

v2.1.0

• Added AI Helper: Generate Release Assistant to create release workflow Markdown from Git history.
• Generates a changelog, deterministic semver bump suggestion, GitHub release notes, npm release summary, PR description, and reviewer checklist.
• Defaults release history to the latest Git tag through HEAD and defaults the target version from package.json when available.
• Opens generated release output in a Markdown review document instead of changing repository files automatically.

v2.0.7

• Fixed provider/model saves so workspace-level settings do not silently override the selected mode.
• Added an Auto-find Codex CLI action in setup that searches PATH and common npm install locations without hardcoding local developer paths.
• Added Check Codex Status to show Codex CLI version, login state, and account label when available.
• Clarified the setup wizard guidance so users know they must save the settings panel to apply provider, model, key, and command choices.
• Improved limits and usage text so truncation, output-token defaults, and tracked token usage are visible in setup.

v2.0.6

• Refreshed bundled fallback model lists and defaults for OpenAI, DeepSeek, Anthropic, Cohere, Gemini, Mistral, OpenRouter, and Hugging Face using current provider model docs and model-list endpoints.
• Updated the default Codex/OpenAI model to gpt-5.5 and switched Gemini, Cohere, Mistral, OpenRouter, and Hugging Face defaults to currently documented model IDs.

v2.0.5

• Fixed the AI Helper Activity Bar settings view contribution so selecting the icon opens the settings webview instead of triggering a missing data provider error.

v2.0.4

• Replaced the AI Helper sidebar action list with a settings-first view: selecting the Activity Bar icon opens the settings panel instead of showing duplicate commands.
• Added a setup wizard mode for first-run configuration and a Run Setup Wizard button for rerunning provider setup later.
• Tightened provider/model validation so non-custom providers only save supported model IDs or models returned by their model API.
• Added model refresh support in the settings UI, including pricing metadata when a provider exposes token billing data.
• Updated DeepSeek defaults to supported deepseek-v4-flash and deepseek-v4-pro model names.

v2.0.3

• Restored public contribution IDs and settings to the codexCommitWidget.* namespace to avoid collisions with the renaming to ai-commit-prompt-helper extension.
• Removed explicit command and view activation events now generated by VS Code from package contributions.
• Added the sidebar view icon required by VS Code package validation.

v2.0.2

• Added a provider-mode setup panel with compact provider/model dropdowns.
• Added SecretStorage-backed API key saving with legacy setting and environment-variable fallbacks.
• Added Hugging Face provider support through its OpenAI-compatible router.
• Added low-cost model presets for OpenRouter, Hugging Face, Mistral, Gemini, and DeepSeek-oriented workflows.

v2.0.1

• Fixed aiCommitPromptHelper.improvePrompt not found by adding explicit command activation events.

v2.0.0

• Rebranded the extension display name to AI Commit & Prompt Helper while keeping the Marketplace extension ID codex-commit-widget.
• Added a provider interface for Codex CLI, OpenAI-compatible APIs, Anthropic Claude, Cohere, Google Gemini, Mistral, DeepSeek, OpenRouter, Hugging Face, and custom OpenAI-compatible endpoints.
• Added AI Helper: Improve Prompt for selected editor text, with review before copy/open/replace.
• Moved settings to aiCommitPromptHelper.*; existing codexCommitWidget.* values were read as fallbacks.

Install

Install as a VS Code extension:

• From Marketplace: search for AI Commit & Prompt Helper
• Or from a .vsix build: Extensions: Install from VSIX...

Quick Start

1. Open a Git repository in VS Code.
2. Select the AI Helper Activity Bar icon.
3. On first use, complete the setup wizard:
   • Choose the provider you want to use.
   • Add the required API key or Codex CLI command.
   • Refresh models when the provider supports model listing.
   • Review or edit the commit-message prompt.
   • Save settings.
4. Stage the files you want to commit.
5. Generate a commit message from the Source Control title button or the Command Palette command AI Helper: Generate Commit Message.
6. The generated message is written directly into the commit message box.

To improve a coding prompt, select prompt text in an editor and run AI Helper: Improve Prompt from the Command Palette. If no text is selected, the extension asks for a prompt, then shows a review document before you copy, open, or replace the improved version.

To prepare release copy, run AI Helper: Generate Release Assistant from the Command Palette. The command reads Git history since the latest tag, asks for the target version, then opens a Markdown document containing a changelog, semver recommendation, GitHub release notes, npm summary, PR description, and reviewer checklist.

Providers

Set codexCommitWidget.provider to one of:

• codexCli (default)
• codexExtensionThenCli
• openai
• deepseek
• anthropic
• cohere
• gemini
• mistral
• openrouter
• huggingface
• customOpenAiCompatible

HTTP providers use provider-specific API keys from VS Code SecretStorage when saved through AI Helper: Open Settings. Existing settings.json API key values and environment variables still work as fallback migration paths.

The settings panel can refresh available models for providers with model-list APIs. It also shows token pricing when the provider returns billing metadata. Fallback model lists are bundled so setup still works if a provider does not expose model listing or a network request fails.

Current routed presets are included for providers that support broad model routing:

• OpenRouter: openai/gpt-5.5, openai/gpt-5.4-mini, google/gemini-3.5-flash, deepseek/deepseek-v4-flash, qwen/qwen3.7-max, mistralai/mistral-medium-3-5
• Hugging Face: openai/gpt-oss-120b, Qwen/Qwen3-Coder-480B-A35B-Instruct, deepseek-ai/DeepSeek-R1, zai-org/GLM-4.5
• Mistral, Gemini, Cohere, DeepSeek, and OpenAI defaults use current provider-documented model IDs.

What It Sends

Commit generation sends staged context only:

• Repository name
• Repository status
• Changed file list
• Diff stats
• Staged patch

Prompt improvement sends only the selected/input prompt text. Absolute local repository paths are not sent.

Release assistant generation sends Git history for the selected release range, changed file names, diff statistics, repository name, package name, and target version. It does not publish a GitHub release, run npm publish, create tags, or modify files.

Settings

Select the AI Helper Activity Bar icon to open the settings panel. On first use, it opens in setup wizard mode. After setup is saved, selecting the icon opens normal settings. Use Run Setup Wizard in the panel if you want to repeat provider setup later.

Each provider mode shows only the fields used by that provider. Non-custom providers use provider-scoped model dropdowns and cannot save arbitrary custom model IDs. The custom model input is reserved for Codex CLI and custom OpenAI-compatible endpoints. API keys entered in the panel are stored in VS Code SecretStorage instead of normal settings.

Core:

• codexCommitWidget.provider
• codexCommitWidget.model
• codexCommitWidget.reasoningEffort
• codexCommitWidget.maxDiffChars
• codexCommitWidget.codexCommand
• codexCommitWidget.codexExtensionCommand
• codexCommitWidget.customOpenAiCompatibleBaseUrl

For Codex CLI modes, use Auto-find Codex CLI in the setup panel to search PATH and common npm install locations, then use Check Codex Status to verify the detected CLI login before saving.

API keys and tokens:

• Setup panel: stores provider keys in VS Code SecretStorage.
• Existing aiCommitPromptHelper.*ApiKey values remain supported as fallbacks.
• Environment fallbacks: OPENAI_API_KEY, DEEPSEEK_API_KEY, ANTHROPIC_API_KEY, COHERE_API_KEY, GEMINI_API_KEY, GOOGLE_API_KEY, MISTRAL_API_KEY, OPENROUTER_API_KEY, HF_TOKEN, HUGGINGFACE_API_KEY, and OPENAI_COMPATIBLE_API_KEY.
• codexCommitWidget.apiKey remains a legacy generic override for the selected HTTP provider.

Prompt/style customization:

• codexCommitWidget.promptTemplate
• codexCommitWidget.additionalPromptInstructions
• codexCommitWidget.temperatureOverride
• codexCommitWidget.topPOverride
• codexCommitWidget.maxOutputTokensOverride

UI and usage:

• codexCommitWidget.statusBarText
• AI Helper: Improve Prompt
• codexCommitWidget.trackTokenUsageAnalytics
• codexCommitWidget.analyticsRetentionDays
• codexCommitWidget.analyticsSummary
• codexCommitWidget.analyticsTotalTokens
• codexCommitWidget.analyticsInputTokens
• codexCommitWidget.analyticsOutputTokens
• codexCommitWidget.analyticsGenerations
• codexCommitWidget.analyticsEstimatedRuns
• codexCommitWidget.analyticsLastUpdated

Detailed examples: docs/configuration.md

Codex CLI

Codex CLI remains supported for local authenticated Codex sessions.

codex login
codex --version
npm install -g @openai/codex@latest

This extension is tuned for Codex CLI 0.120.0 and newer.

Token Usage Analytics

Token usage is stored in settings and auto-pruned after analyticsRetentionDays (default 7 days). Providers that do not return usage metadata fall back to an estimate.

Troubleshooting

• No repository detected: open a folder/workspace with a Git repo.
• No staged changes: stage files before generating.
• Codex command not found: open the setup wizard or set codexCommitWidget.codexCommand.
• HTTP provider auth errors: save the key in AI Helper: Open Settings, or set the matching environment variable.
• Wrong model for provider: open settings, refresh models, and save a model from that provider's dropdown.
• Legacy settings: v2 reads existing aiCommitPromptHelper.* values as fallbacks, but new settings should use codexCommitWidget.*.

License

MIT