Skip to content

Security: ranxianglei/acp-kernel

SECURITY.md

Security Policy

Supported versions

acp-kernel is pre-1.0 and releases often. Security fixes land on master and ship in the next npm release. Only the latest published version is supported.

Version Supported
latest 0.0.x on npm ✅
older releases, pr-* npm tags ❌

Reporting a vulnerability

Please do not open a public issue or pull request for a security vulnerability.

Report it privately through GitHub: Security → Report a vulnerability.

Please include:

  • the affected version or commit,
  • the affected module (e.g. src/ccr.ts, src/wire/openai.ts),
  • a description of the issue and its impact,
  • steps or a minimal input to reproduce it, and
  • any suggested fix, if you have one.

What to expect

  • An acknowledgement within 7 days.
  • An initial assessment (accepted or declined, with a severity) within 14 days.
  • For accepted reports, a fix and a GitHub Security Advisory, with a CVE requested where appropriate. Reporters are credited unless they ask not to be.

Please give us a reasonable window to release a fix before any public disclosure. We aim for 90 days or less.

Scope

acp-kernel never calls a model or the network itself, but it processes text that is not trusted: tool results, model output, and conversation history supplied by the host. Examples of issues that are in scope:

  • Untrusted content (tool results, fetched pages, model-written tool arguments) being promoted to a more privileged role or channel, such as a system or developer message, or escaping its delimiters.
  • Denial of service from crafted input, e.g. catastrophic regex backtracking or unbounded memory or CPU use in parsing, compression, search, or wire encoding.
  • Protected-content filtering, message-ref, or block-state logic that can be bypassed to hide, alter, or misattribute conversation content.
  • Path traversal or unsafe file writes in acp-kernel/persist.
  • Weaknesses in this repository's CI and release workflows (.github/workflows/) that could lead to a compromised npm package.

Out of scope:

  • Model behaviour on its own, such as a model obeying a prompt injection that acp-kernel delivered at its correct, unprivileged role.
  • Vulnerabilities in host adapters (e.g. pai-acp) or agent platforms. Please report those to their own projects.
  • Vulnerabilities in development-only dependencies that do not affect the published package.

There aren't any published security advisories