Angrist executes code repair in local environments, manages Git worktrees, and interfaces with language models. We take security and repository integrity seriously.

Security updates and patches are actively applied to the following versions:

If you discover a potential security vulnerability in Angrist (such as an AST scope escape, unintended file execution, or command injection), please report it responsibly.

Do not open a public GitHub issue for security vulnerabilities.

Instead, please send an email to: security@angrist.org (or contact the primary maintainer directly via GitHub Security Advisories).

Please include the following details in your report:

1. Description of the vulnerability and its potential impact.
2. Step-by-step reproduction instructions or a minimal proof-of-concept repository.
3. Your operating system and Python version.
4. Any proposed mitigations or patches, if available.

• Initial Response: You will receive an acknowledgment of your report within 48 hours.
• Triage & Assessment: The maintainers will evaluate the vulnerability and determine severity.
• Resolution & Release: A fix will be developed, reviewed, and tested in a private branch before a public security release is published.
• Credit: We will publicly credit your contribution in release notes (unless you prefer to remain anonymous).

Angrist is designed with defense-in-depth principles:

1. AST-Enforced Isolation: Candidate patches cannot modify anything outside the target node. Top-level script executions, arbitrary module imports, and external file writes via the patch payload are blocked at the AST verification layer.
2. Worktree Isolation: Patches and tests execute inside disposable worktree sandboxes outside the main repository tree. An untrusted patch cannot inadvertently overwrite unstaged working changes in your active branch.
3. No Automatic Shell Execution of Unsanitized Inputs: File paths and target names are verified and sanitized before subprocess invocation.