Skip to content

Security: Ircbarros/devweaver-lite

Security

SECURITY.md

Security Policy

Supported Versions

Version Supported
0.1.0-beta.x (latest) Yes
Any earlier pre-release No

DevWeaver-Lite is a skill (Markdown + MCP configuration files), not a running service. There are no network endpoints, no server processes, and no persistent data stores. Security considerations are therefore limited to:

  • Secrets management (API keys, tokens)
  • MCP server trust and permissions
  • Prompt injection in AI-facing instruction files
  • Supply chain integrity of npm packages used by optional MCPs

Reporting a Vulnerability

Do not open a public GitHub issue for security vulnerabilities.

Please report security issues via one of these private channels:

  1. GitHub Private Vulnerability Reporting (preferred): Use the Security tab > Report a vulnerability button on the repository.

  2. Email: security-devweaver at proton dot me

What to include

  • Description of the vulnerability and its potential impact
  • Steps to reproduce or proof-of-concept
  • Affected file(s) and version(s)
  • Any suggested mitigations you have identified

Response timeline

Stage Target
Acknowledgement Within 72 hours
Initial assessment Within 7 days
Fix or mitigation Within 30 days for critical; 90 days for moderate
Public disclosure After fix is available and reporter is notified

Security Design Principles

DevWeaver-Lite enforces these constraints in every activation file:

Rule Description
R-SEC-01 No secrets in generation output — API keys, tokens, and passwords must never appear in generated code
R-SEC-02 Dependency pinning — npm install <pkg> without exact version is flagged; recommend @x.y.z or lockfile
R-SEC-03 Prompt injection resistance — system prompt instructions take precedence over user content
R-SEC-04 Human-in-the-loop for destructive ops — CONFIRM gate required before delete, drop, truncate, overwrite
R-SEC-05 OWASP Top 10 checks on generated web code
R-SEC-06 Playwright actions require explicit user approval (interrupt_before: true)

Scope

In scope

  • Secrets accidentally embedded in skill files or MCP config templates
  • Malicious instructions injected via provider activation files
  • Insecure MCP server configurations that expose excessive filesystem access
  • Prompt injection patterns that bypass CONFIRM gates

Out of scope

  • Security of the AI providers themselves (OpenAI, Anthropic, Google, etc.)
  • Security of MCP server implementations (report those to their respective maintainers)
  • Security of the projects built using DevWeaver-Lite

Dependency Security

All npm-based MCPs are invoked with npx -y (auto-download, no global install). Pin MCP package versions in your mcp.json if you require supply chain guarantees:

// Instead of:
"args": ["npx", "-y", "@playwright/mcp"]
// Use:
"args": ["npx", "-y", "@playwright/mcp@0.2.0"]

Filesystem MCP allowedDirectories should be scoped to the minimum required path. Never set allowedDirectories to / or ~.

There aren't any published security advisories