Bug Agent

A simple GitHub issue analysis tool that fetches GitHub issues, summarizes the issue content, and asks a local language model for advice or next steps.

What this agent does

• Fetches an issue from a GitHub repository
• Converts issue metadata into a structured internal format
• Optionally enriches analysis with comments, related issues, repository metadata, and top contributors
• Builds a prompt containing issue details and selected context
• Sends the prompt to a local Hugging Face causal LM for advice, or uses deterministic triage with --no-model
• Saves analysis results to a local JSON file for later reference

Current implementation

The current code path includes:

1. src/main.py

   • parses command-line options
   • loads the GitHub client and the issue advisor
   • fetches either a specific issue or the latest open issue
   • optionally fetches comments, related issues, contributors, and repo metadata
   • summarizes and formats the issue context
   • calls the advisor to generate natural-language guidance
   • writes the result into bug_agent_memory.json

2. src/github_client.py

   • uses the GitHub REST API to fetch issues
   • supports retrieving one issue, the latest open issue, issue comments, issue search, contributors, and repository metadata
   • skips pull requests when enumerating issues

3. src/summarizer.py

   • converts raw GitHub issue JSON into a structured object
   • builds a related-issue search query from issue-specific terms
   • formats issue context into prompt text for the LM

4. src/issue_advisor.py

   • loads a Hugging Face causal LM model and tokenizer
   • constructs a prompt with a system instruction and issue details
   • generates advice text from the model output
   • falls back to deterministic triage when the model is disabled or unavailable

Requirements

• Python 3.10+
• requests
• transformers

Install dependencies with:

pip install -r requirements.txt

Quick Start

Optionally set a custom model name:

export MODEL_NAME=HuggingFaceTB/SmolLM-1.7B

Run the agent for the default repository and latest open issue:

python -m src.main

Run a fast deterministic analysis without loading a Hugging Face model:

python -m src.main --no-model

If model-backed analysis is slow on CPU, limit generation length:

python -m src.main --max-new-tokens 80

Run the agent for a specific issue number:

python -m src.main --issue 44593

Run the agent for a different repository:

python -m src.main --repo owner/repo

Include richer GitHub context:

python -m src.main --repo owner/repo --issue 123 --comments 5 --related 5 --contributors 3 --repo-context

Print the full structured result as JSON:

python -m src.main --issue 44593 --comments 3 --related 3 --no-model --json

Run the test suite:

python -m unittest discover

CLI options

• --comments N: include up to N issue comments.
• --related N: search for up to N potentially related issues.
• --contributors N: include the top N repository contributors.
• --repo-context: include repository metadata such as language and default branch.
• --no-model: skip Hugging Face model loading and use deterministic triage.
• --max-new-tokens N: limit Hugging Face generation length. Lower values return faster.
• --memory-file PATH: write persistent analysis memory to a custom JSON file.
• --json: print the structured issue context and response as JSON.
• --verbose: enable debug logging.

Output

The agent prints the selected issue URL and analysis response, and stores the structured issue context plus response in bug_agent_memory.json.