Skip to content

Security: dipjyotimetia/deep-agent-action

SECURITY.md

Security Policy

Reporting a vulnerability

Please do not report security vulnerabilities through public GitHub issues, discussions, or pull requests.

Instead, report privately using one of:

Please include, where possible:

  • A description of the issue and its impact.
  • Steps to reproduce or a proof of concept.
  • Affected version / commit and configuration.
  • Any suggested remediation.

We will acknowledge your report, keep you updated on progress, and credit you in the advisory once a fix is released (unless you prefer to remain anonymous).

Scope

This action runs a model-driven agent that executes commands in your runner. Reports we're especially interested in:

  • Bypasses of the authorization checks (permission gating, human-actor check).
  • Bypasses of fork-PR protection or the shell command allow/deny guardrails.
  • Secret exposure — any way provider keys, the GITHUB_TOKEN, or the GitHub App key reach the agent's shell or logs.
  • Token/privilege escalation via the minted GitHub App installation token.

Please review the security model first — several behaviors that look risky (e.g. the agent running shell commands) are intentional and bounded by guardrails. The model is a layered guardrail, not a sandbox; running untrusted instructions on an untrusted provider is out of scope.

Supported versions

This project is pre-1.0. Security fixes are applied to the latest main. Pin to a commit SHA for reproducible deployments (see Versioning).

Best practices for operators

For operator hardening guidance, see the hardening checklist in the security model.

There aren't any published security advisories