Skip to content

Security: openclaw/acpx

Security

SECURITY.md

Security Policy

If you believe you have found a security issue in acpx, please report it privately.

Reporting

Report vulnerabilities for this repository at:

If you are unsure whether the issue belongs in acpx, email security@openclaw.ai and include:

  1. Title
  2. Severity assessment
  3. Impact
  4. Affected component
  5. Technical reproduction
  6. Demonstrated impact
  7. Environment
  8. Remediation advice

Reports without reproduction steps, demonstrated impact, and remediation advice may be deprioritized. Given the volume of AI-generated scanner findings, we must ensure we're receiving vetted reports from researchers who understand the issues.

Bug Bounties

acpx is a labor of love. There is no bug bounty program and no budget for paid reports. Please still disclose responsibly so we can fix issues quickly. The best way to help the project right now is by sending PRs.

Maintainers: GHSA Updates via CLI

When patching a GHSA via gh api, include X-GitHub-Api-Version: 2022-11-28 (or newer). Without it, some fields, notably CVSS, may not persist even if the request returns 200.

Scope

acpx is a local, headless CLI client for the Agent Client Protocol (ACP). It runs on a trusted machine, spawns local ACP adapters and agents, and stores session/config state on disk.

Security issues in scope generally include:

  • unintended command execution caused by acpx
  • unsafe handling of local credentials or auth material configured through acpx
  • path traversal or filesystem boundary bypasses in acpx client features
  • permission-policy bypasses in fs/* or terminal/* client method handling
  • leakage of sensitive local data through acpx session persistence or output modes

Out of Scope

The following are usually out of scope for this repository:

  • vulnerabilities in upstream coding agents, ACP adapters, or third-party CLIs that acpx launches
  • issues that require prior write access to trusted local state such as ~/.acpx/, project files, or shell startup files
  • prompt injection by itself, unless it demonstrates a concrete acpx security boundary bypass
  • insecure local machine administration or multi-user host setups where the OS trust boundary is already lost
  • use of unrecommended or intentionally unsafe custom agent commands provided through --agent

If the issue is actually in an upstream tool, please report it to that project. Examples include:

Trust Boundaries

acpx assumes the local machine and user account running it are trusted.

  • Global config is stored in ~/.acpx/config.json.
  • Session metadata and history are stored under ~/.acpx/sessions/.
  • Project config may be read from <cwd>/.acpxrc.json.
  • Spawned adapters and agents run with the privileges of the current user.

If an attacker can already modify those files or the commands that acpx launches, they have already crossed the primary trust boundary.

Operational Guidance

  • Keep acpx, Node.js, and the underlying coding agents up to date.
  • Review any custom commands configured through --agent or config.agents.*.command before using them.
  • Treat ~/.acpx/config.json as sensitive if it contains auth credentials.
  • Do not share session files or command output if they may contain prompts, file paths, or credentials from local work.
  • Prefer running acpx on a trusted local machine or isolated CI runner.

Runtime Requirements

acpx requires Node.js 22.12.0 or later.

Verify your version with:

node --version

There aren’t any published security advisories