Skip to content

Security: specfuse/orchestrator

Security

SECURITY.md

Security policy

Thank you for helping keep the Specfuse Orchestrator and its downstream consumers secure.

Reporting a vulnerability

Please do not report security vulnerabilities through public GitHub issues, discussions, or pull requests. Public reports give attackers a head start while a fix is being prepared.

Instead, open a private security advisory on this repository. The advisory is visible only to you and the maintainers until a fix is published.

Please include:

  • A description of the vulnerability and its potential impact.
  • Steps to reproduce it (proof-of-concept welcome).
  • Affected files / agents / scripts (e.g., a specific skill, a shell helper, a schema).
  • Whether the issue is in scaffolding (this repo's content) or could affect a downstream's coordination state (/features/, /events/, /inbox/, override registry).

We aim to acknowledge reports within 7 days and provide a remediation timeline within 14 days.

Scope

In scope:

  • Vulnerabilities in the orchestrator's scaffolding: the specfuse pm CLI (init / upgrade), schemas, agent configurations that could lead to prompt injection, escalation bypass, or unauthorized writes outside documented surfaces.
  • Documentation that could mislead an operator into an insecure setup (missing never-touch.md boundary, unsafe gh permission scopes, etc.).

Out of scope:

  • Issues in downstream private orchestration repos (those are the downstream operator's responsibility).
  • Issues in Claude Code itself, the Anthropic API, or upstream tools — please report those to the relevant project.
  • Issues in the Specfuse generator or _generated/ content in component repos — those belong to the Specfuse generator project.

Disclosure timeline

After a fix lands, we'll publish an advisory describing the issue and credit the reporter (if desired). Until the fix is public, please keep details private.

Existing safeguards worth knowing about

  • shared/rules/never-touch.md enumerates filesystem paths agents must never read or write (secrets, generated directories, /business/ in product specs, branch protection settings).
  • shared/rules/security-boundaries.md documents what agents may and may not exfiltrate or transmit.

There aren't any published security advisories