Skip to content

Security: patterson-agents/actdata-plugins

SECURITY.md

Security Policy

Reporting a vulnerability

Report suspected vulnerabilities in actdata-plugins privately, never through a public GitHub issue.

Use GitHub's private vulnerability reporting on this repository ("Security" tab, then "Report a vulnerability") if it is enabled. That flow creates a private draft advisory visible only to maintainers and gives you a secure channel to attach reproduction detail.

If it is not enabled here, report through Patterson's internal security contact.

Caution

[TBD: the Patterson-wide security contact or mailbox has not been recorded in this repository.] Do not report a vulnerability publicly while this is unresolved — escalate through your normal Patterson security channel instead.

Please include, where you can:

  • The affected plugin, skill, agent, command, or script.
  • Reproduction steps or a minimal example.
  • The potential impact as you understand it — what an attacker could actually do, not just what looks unusual.

What is in scope

  • The plugin manifests, skills, agents, and commands under plugins/.
  • .claude-plugin/marketplace.json — in particular anything that would cause an agent to install or trust something unintended.
  • The validator scripts under scripts/ and the CI configuration under .github/.
  • This repository's supply chain. It currently has a near-empty dependency footprint, and the two validators in scripts/ deliberately import only node:* builtins so the gate itself is not a supply-chain surface.

Particular attention: prompt content is executable

A plugin in this repository is mostly instructions to an agent. A malicious or careless edit to a SKILL.md, an agent system prompt, or a slash command is not inert text — it changes what an agent will do on someone else's machine, with that person's credentials.

Treat the following as security-relevant, not merely stylistic:

  • Instructions that tell an agent to exfiltrate file contents, environment variables, or credentials to any network destination.
  • Instructions that suppress confirmation prompts, or that encourage running commands without showing them to the user.
  • A hook that blocks or rewrites tool calls without a documented off switch.
  • An MCP server configured over plain HTTP or WS rather than HTTPS or WSS.
  • An absolute path where ${CLAUDE_PLUGIN_ROOT} belongs — it can point at a directory the plugin author did not intend on a different machine.

Dependencies

Before adding or upgrading any third-party package:

socket package shallow npm pkg:npm/<name>@<version> --markdown

Flag anything scoring under 90 on any of the five dimensions (supply chain, maintenance, quality, vulnerability, license) and get confirmation before installing. Read the [high]/[middle]/[low] alerts line as well.

Out of scope for this file

  • Vulnerabilities in ACT Data products or infrastructure that a plugin here happens to describe. Report those through the normal channel for that system.
  • Upstream defects in Anthropic's plugin-dev, from which plugins/act-plugin-dev/ is forked. Report those to Anthropic. If a defect has security impact as vendored here, report it here as well so the fork can be patched independently.

Response

This repository does not commit to a response-time service level agreement. [TBD: no vulnerability-response SLA has been set for this repository.] Reports are triaged by the owning team in CODEOWNERS for the affected path.

There aren't any published security advisories