Skip to content

Security: zakelfassi/skills-driven-development

Security

SECURITY.md

Security Policy

Supported versions

SkDD is an early-stage open methodology + CLI + VS Code extension scaffold. Only the latest minor version on main is supported for security fixes. Older tagged releases get best-effort patches if the fix is small, but no guarantees.

Version Supported
0.3.x
0.2.x best-effort
< 0.2

Reporting a vulnerability

Please do not open a public GitHub issue for security vulnerabilities.

Use GitHub's Private Vulnerability Reporting instead:

  1. Go to https://github.com/zakelfassi/skills-driven-development/security/advisories
  2. Click Report a vulnerability
  3. Describe the issue — what you observed, how to reproduce, and your threat model
  4. If you have a suggested fix, include it (or link a private fork)

You can also reach the maintainer directly at zakelfassi+security@gmail.com. PGP is not currently required.

What counts as a vulnerability

SkDD's threat surface is small — it's a CLI that reads + writes files in your project, a JSON Schema, a set of markdown templates, and a VS Code extension that shells out to the CLI. Things we consider in-scope:

  • Command injection in the CLI: any skdd subcommand that runs a shell with user-supplied input without escaping
  • Path traversal: skdd init, skdd forge, skdd link, skdd import writing outside the project root
  • Symlink attacks: skdd link following a malicious symlink to clobber user data outside skills/ or a configured mirror path
  • Arbitrary file overwrite without --force: any command that silently replaces user content the user didn't explicitly opt into replacing
  • Supply-chain issues in cli/ dependencies (we pin via pnpm-lock, but upgrades can introduce risk)
  • VS Code extension issues: the extension shells out to skdd via vscode.window.createTerminal or vscode.ShellExecution — anything that lets a crafted workspace file execute unintended commands is in-scope

Out of scope:

  • Skill content: skills are markdown files the user authors or receives. If a skill says "run rm -rf /", that's social engineering of the user, not a SkDD bug. skdd validate will not analyze skill content for semantic safety.
  • Harness behavior: how Claude Code / Codex / Cursor / etc. execute a skill is outside SkDD's control. Report those to the respective vendors.
  • DoS via large files: SkDD doesn't run in production and has no notion of request rate. If skdd validate is slow on a huge colony, open a normal issue.
  • Publishing / marketplace submissions: SkDD does not yet publish or submit colonies automatically.

Response timeline

  • 72 hours for acknowledgement (maintainer will reply via the advisory thread)
  • 14 days for an initial patch plan for critical issues
  • 30 days for a public fix + advisory on non-critical issues
  • 90 days maximum embargo before public disclosure, unless we agree a longer window in writing

If the issue is actively exploited in the wild, we will disclose faster than 30 days.

Credit

Researchers who report valid vulnerabilities will be credited in the advisory and CHANGELOG.md under the fix, unless they request anonymity. There is no paid bug bounty.

See also

  • docs/spec/colony-v1.json — JSON Schema for .colony.json
  • CONTRIBUTING.md — how to submit non-security patches and improvements
  • ROADMAP.md — a security doc with signing conventions and Snyk integration is planned (P4.6 in the project plan)

There aren’t any published security advisories