You can run the mini-swe-agent locally to test PRs before deploying to GitHub Actions.

1. Install dependencies:

   npm install
   npm run bundle  # Build the CLI

2. Set up required environment variables:

   # GitHub authentication (choose one):
   export GITHUB_TOKEN=$(gh auth token)  # If using gh CLI
   # OR
   export GITHUB_TOKEN=ghp_your_token_here
   
   # Model API keys (at least one required):
   export ANTHROPIC_API_KEY=your_anthropic_key
   # OR
   export OPENAI_API_KEY=your_openai_key

Test the agent without creating PRs or comments on GitHub:

npm run cli -- --repo owner/repo --pr 123 --dry-run

This will:

• Clone the PR repository
• Run the mini-swe-agent to analyze and fix issues
• Display the generated patch locally
• Not make any changes to GitHub

Run the agent and create actual PRs or inline comments on GitHub:

npm run cli -- --repo owner/repo --pr 456

This will:

• Clone the PR repository
• Run the mini-swe-agent
• Create a follow-up PR or post inline comments based on the agent's decision

If you have a specific workflow run that failed, you can provide its ID:

npm run cli -- --repo owner/repo --pr 789 --workflow-run-id 12345

-r, --repo <owner/repo>          Repository in "owner/repo" format
-p, --pr <number>                Pull request number (required)
-d, --dry-run                    Show patch locally without PRs/comments
-t, --token <token>              GitHub token (default: GITHUB_TOKEN or gh)
-w, --workflow-run-id <id>       Workflow run ID for failure logs
-o, --output-dir <path>          Directory for debug artifacts
--clickhouse-url <url>           ClickHouse URL for tracking
--clickhouse-table <name>        ClickHouse table name
-c, --cost-limit <dollars>       Cost limit (default: 3.0)
--timeout <minutes>              Timeout in minutes (default: 30)
-h, --help                       Show help message

# Dry run on a public repository
npm run cli -- --repo tensorzero/tensorzero --pr 100 --dry-run

# Run on your own repository with custom settings
export GITHUB_TOKEN=$(gh auth token)
npm run cli -- \
  --repo myorg/myrepo \
  --pr 42 \
  --cost-limit 5.0 \
  --timeout 45 \
  --output-dir ./debug-output

# Analyze a specific failed workflow run
npm run cli -- \
  --repo owner/repo \
  --pr 123 \
  --workflow-run-id 9876543210

• npm install
• npm run bundle will build the action for distribution.

• npm run bundle will build the action.
• Prepare a ClickHouse Cloud database
  • We need to add a table for Inference => PR association; assuming one inference per PR. Might be able to iterate.
  • Configure the action inputs clickhouse-url, clickhouse-table, and optional authentication parameters so the action can write inference to PR mappings after creating follow-up pull requests.
• Configure secrets for the repository:
  • CI_BOT_OPENAI_API_KEY: OpenAI API key, used when starting TensorZero gateway
  • CI_BOT_CLICKHOUSE_URL: ClickHouse URL for both TensorZero gateway and the GitHub PR to inference mapping; expected format is http[s]://[username:password@]hostname:port[/database].
• Configure GitHub Actions permissions for the repository:
  • Under "Settings > Actions > General", check the box for "Allow GitHub Actions to create and approve pull requests".

We need to create a new table to store GitHub PR to Inference Map:

CREATE TABLE GitHubBotPullRequestToInferenceMap
(
   pull_request_id UInt128,
   inference_id String,
   episode_id String,
   created_at DateTime DEFAULT now(),
   original_pull_request_url String
)
ENGINE = ReplacingMergeTree
ORDER BY pull_request_id;

• Long context / large PRs

This project includes a helper script, script/release designed to streamline the process of tagging and pushing new releases for GitHub Actions.

GitHub Actions allows users to select a specific version of the action to use, based on release tags. This script simplifies this process by performing the following steps:

1. Retrieving the latest release tag: The script starts by fetching the most recent SemVer release tag of the current branch, by looking at the local data available in your repository.
2. Prompting for a new release tag: The user is then prompted to enter a new release tag. To assist with this, the script displays the tag retrieved in the previous step, and validates the format of the inputted tag (vX.X.X). The user is also reminded to update the version field in package.json.
3. Tagging the new release: The script then tags a new release and syncs the separate major tag (e.g. v1, v2) with the new release tag (e.g. v1.0.0, v2.1.2). When the user is creating a new major release, the script auto-detects this and creates a releases/v# branch for the previous major version.
4. Pushing changes to remote: Finally, the script pushes the necessary commits, tags and branches to the remote repository. From here, you will need to create a new release in GitHub so users can easily reference the new tags in their workflows.

This template includes a GitHub Actions workflow, licensed.yml, that uses Licensed to check for dependencies with missing or non-compliant licenses. This workflow is initially disabled. To enable the workflow, follow the below steps.

1. Open licensed.yml

2. Uncomment the following lines:

   # pull_request:
   #   branches:
   #     - main
   # push:
   #   branches:
   #     - main

3. Save and commit the changes

Once complete, this workflow will run any time a pull request is created or changes pushed directly to main. If the workflow detects any dependencies with missing or non-compliant licenses, it will fail the workflow and provide details on the issue(s) found.

Whenever you install or update dependencies, you can use the Licensed CLI to update the licenses database. To install Licensed, see the project's Readme.

To update the cached licenses, run the following command:

licensed cache

To check the status of cached licenses, run the following command:

licensed status