We take security vulnerabilities seriously. If you discover a security issue, please report it responsibly.

Please use GitHub's private vulnerability reporting feature:

1. Go to the Security tab of this repository
2. Click "Report a vulnerability"
3. Fill out the vulnerability report form

This ensures your report is kept confidential until a fix is available.

• A description of the vulnerability
• Steps to reproduce the issue
• Potential impact of the vulnerability
• Any suggested fixes (optional)

• Acknowledgment: We will acknowledge receipt of your report within 48 hours
• Assessment: We will investigate and provide an initial assessment within 7 days
• Resolution: We aim to release a fix within 30 days for critical vulnerabilities
• Disclosure: We will coordinate with you on public disclosure timing

This security policy covers:

• The Tara Code CLI application
• Official distribution channels (GitHub releases, Homebrew tap)

taraengine reports go to its own policy: https://github.com/tara-vision/taraengine/blob/main/SECURITY.md

• Vulnerabilities in third-party dependencies (please report to the upstream project)
• Issues with third-party self-hosted LLM servers (vLLM, llama.cpp)
• Social engineering attacks

Every release ships checksums.txt, a keyless cosign signature of it as a Sigstore bundle (checksums.txt.sigstore.json), and SLSA provenance.

VERSION=v2.1.0
BASE=https://github.com/tara-vision/taracode/releases/download/${VERSION}
curl -fsSLO ${BASE}/checksums.txt -O ${BASE}/checksums.txt.sigstore.json
cosign verify-blob --bundle checksums.txt.sigstore.json \
  --certificate-identity-regexp 'https://github.com/tara-vision/taracode/.github/workflows/release.yml@refs/tags/v.*' \
  --certificate-oidc-issuer https://token.actions.githubusercontent.com checksums.txt
sha256sum --ignore-missing -c checksums.txt   # shasum -a 256 -c on macOS

When using Tara Code:

• Keep your installation updated to the latest version
• Review commands before execution when using the shell tool, and review the policy before switching to operate mode
• Be cautious with file operations in sensitive directories
• Ensure your self-hosted LLM servers (taraengine, vLLM, llama.cpp) are properly secured if exposed to a network
• .taracode/audit.jsonl stores the model's raw command for every mutation, so a secret it echoes into a command (rather than one taracode reads from a file or an environment variable) lands unredacted in that 0600 file; treat it like any other local file with your command history in it