codehawk

Codehawk is a tool that uses AI (via OpenRouter) to analyze and triage GitHub issues.

Setup

An OpenRouter API key is required. The OpenRouter API key is expected at the ~/.openrouter/key path.

mkdir -p ~/.openrouter
echo "your-openrouter-api-key" > ~/.openrouter/key

A GitHub token can be used for authenticated requests to the GitHub API, which can help avoid rate limiting. If you have a personal access token, store it in ~/.github/token:

mkdir -p ~/.github
echo "your-github-token" > ~/.github/token

Usage

Analyze recent issues and pull requests

To get an AI-generated summary of issues and pull requests created or updated in one or more repositories within a specific timeframe (defaulting to the last 7 days):

codehawk analyze [--days <number_of_days>] <owner/repo> [<owner/repo> ...]

--days <number_of_days>: Optional. Specifies the number of past days to fetch issues and pull requests from. Defaults to 7.

<owner/repo>: The GitHub repository path(s) (e.g., rust-lang/rust). You can specify multiple repositories.

codehawk analyze --days 14 owner1/repoA owner2/repoB

Example: Analyze issues and pull requests in my-org/my-project from the last 7 days (using default):

codehawk analyze my-org/my-project

Prioritize recent issues and pull requests

To get an AI-generated prioritization of issues and pull requests for more repositories within a specific timeframe (defaulting to the last 7 days):

codehawk prioritize [--days <number_of_days>] <owner/repo> [<owner/repo> ...]

--days <number_of_days>: Optional. Specifies the number of past days to fetch issues and pull requests from. Defaults to 7.

<owner/repo>: The GitHub repository path(s) (e.g., rust-lang/rust). You can specify multiple repositories.

codehawk prioritize --days 14 owner1/repoA owner2/repoB

Example: Prioritize issues and pull requests in my-org/my-project from the last 7 days (using default):

codehawk prioritize my-org/my-project

Triage a specific issue

To get an AI-generated triage report, potentially including a minimal reproducer, for a specific issue:

codehawk triage <owner/repo> <issue_number>

Review a pull request

To get a review for a pull request

codehawk review <owner/repo> <pull_request_number>

Chat

To start an interactive chat session with the AI:

codehawk chat

Within the chat session, you can use the following commands:

• \quit: Exit the chat session.
• \clear: Clear the chat history.
• \limit N: Reduces the history of messages to the N most recent messages. For example, \limit 10 keeps only the last 10 messages.
• \backtrace N: Removes the last N messages from the history. For example, \backtrace 5 removes the last 5 messages, effectively going back 5 steps.
• \show: Displays the current chat history, with messages numbered and showing the role (user/assistant) and content.

List available models

To list the AI models available through OpenRouter:

codehawk models

This command fetches the list of models from https://openrouter.ai/api/v1/models and displays their IDs.

Options

You can customize the AI interaction using global options placed before the command:

--model <model_name>: Specify a different model available on OpenRouter (Default: google/gemini-2.5-pro).

--max-tokens <number>: Set the maximum number of tokens for the AI's response (Default: 16384).

--parameter <name>=<value>: Set model parameters to control AI behavior. Can be used multiple times.

--no-system-prompts: Skip adding any system prompts to the conversation.

Model Parameters

Use --parameter to fine-tune the AI model's behavior. You can specify multiple parameters:

# Control creativity/randomness (0.0 = deterministic, 1.0 = very creative)
codehawk --parameter temperature=0.7 chat

# Control response diversity (nucleus sampling)
codehawk --parameter top_p=0.9 chat

# Limit token selection to top K choices
codehawk --parameter top_k=40 chat

# Reduce repetitive text
codehawk --parameter frequency_penalty=0.3 chat

# Encourage new topics/ideas
codehawk --parameter presence_penalty=0.6 chat

# Alternative repetition control
codehawk --parameter repetition_penalty=1.1 chat

# Set minimum probability threshold
codehawk --parameter min_p=0.05 chat

# Adaptive sampling
codehawk --parameter top_a=0.2 chat

# Set seed for reproducible outputs
codehawk --parameter seed=12345 chat

Parameter Examples by Use Case

Creative Writing (more random and diverse):

codehawk --parameter temperature=0.8 --parameter top_p=0.95 --parameter presence_penalty=0.6 prompt "Write a story"

Code Generation (more focused and deterministic):

codehawk --parameter temperature=0.2 --parameter top_p=0.9 --parameter frequency_penalty=0.1 prompt "Write a Python function"

Analysis Tasks (balanced and consistent):

codehawk --parameter temperature=0.3 --parameter top_p=0.8 analyze --days 7 owner/repo

Parameter Types

• Numbers: 0.7, 40, 12345 (automatically detected)
• Booleans: true, false
• Strings: "text" or text (for non-numeric values)
• Null: null

Note: Parameter availability depends on the model being used. Check your model's documentation for supported parameters.

Use a raw query

codehawk prompt file1 [file2 ....]

It allows to pass a raw request to the AI model.

License

codehawk is licensed under the GNU General Public License v2.0 or later.