Skip to content

Security: frankxai/Starlight-Intelligence-System

Security

SECURITY.md

Security Policy — Starlight Intelligence System

Reporting a vulnerability? Don't open a public issue. Use one of the private channels below — we'll respond within 72 hours.


Why this repo needs a security policy

SIS is the substrate + reference operational layer for the Starlight Intelligence Protocol. The code in this repo handles:

  • Secrets at rest — Infisical workspace tokens, AI provider API keys (Vercel, Anthropic, Stripe), cloud account credentials. Per the privacy split (Tier 1 cleanup, 2026-05-11), real secrets live in operator-local private/ (gitignored) — but the loaders, attestation flows, and key rotation paths are public surface.
  • Financial stateprivate/business-registry.json carries cash positions, runway calculations, and per-entity P&L. The src/finance/ reference build composes these into /yolo session-open payloads.
  • Cross-system orchestration/yolo Hive sessions drive multi-repo git operations under operator authority. A compromised orchestrator can ship to production, push to remote, or merge alliance-touched repos against sovereignty rules.
  • SIP attestationBuilt on SIP attestations are non-trivial to forge but not impossible. Misuse could undermine the social-layer contract that the protocol depends on.

We treat security issues here as substrate-class — same seriousness as sovereignty-clause violations.


Reporting channels (in order of preference)

1. GitHub Security Advisories (recommended)

Open a private advisory at: https://github.com/frankxai/Starlight-Intelligence-System/security/advisories/new

This is the preferred channel because it:

  • Stays private until you and we agree to publish
  • Lets us coordinate a fix branch privately
  • Issues CVE numbers automatically if assigned

2. Email

If GitHub Advisories isn't an option: security@starlightintelligence.org (PGP-friendly — request the public key in your initial message if you want to encrypt).

Backup: open a GitHub issue titled Security report — please contact me without disclosing the vulnerability; we'll reach out via the email on your GitHub profile.


What to include in your report

Make our response fast:

  1. Affected versions / commit SHAs — the closer to a specific tag the better.
  2. Impact — what an attacker can do (read secrets, escalate to remote code execution, bypass attestation, etc.).
  3. Reproduction — minimal steps. If a PoC requires a payload or crafted file, attach it (or describe it textually).
  4. Suggested fix — optional, but appreciated.
  5. Disclosure preferences — coordinated disclosure timeline, whether you want credit, whether you want a CVE.

Our response commitment

  • 72 hours: initial acknowledgement (often much faster).
  • 7 days: triage verdict — confirmed / not-reproducible / duplicate / out-of-scope, with rationale.
  • 30 days: fix shipped for confirmed issues, OR a detailed update on why it's taking longer.
  • 90 days: default coordinated-disclosure timeline. We'll work with you on shorter (if actively exploited) or longer (if a complex fix is in progress) timelines.

In-scope

Issues we want to hear about:

  • Secret leakage paths — committed files exposing .env content, tokens, cash positions, or operator-private state.
  • Privilege escalation in the MCP server, /yolo Hive conductor, or /starlight-board substrate gate.
  • Path traversal, command injection, or unsafe deserialization in the reference operational layer.
  • Bypasses of the sovereignty clause (alliance_touched: true repos being operated on by /yolo despite the exclusion rule).
  • Forged or stripped Built on SIP attestations.
  • Supply-chain risks — typosquatting, dependency confusion, lockfile tampering vectors against @arcanea/starlight-intelligence-system.
  • Privacy-split leaks — operator-private private/ state reaching committed surface.

Out-of-scope

We'll respond politely but won't treat these as security issues:

  • Bugs that require operator-level access to begin with (the operator can already do anything they want on their own machine).
  • Social-engineering of the operator outside of the protocol surface (we can't fix non-protocol attacks).
  • Issues in forks of this repo — please report those to the fork maintainer.
  • Issues in Arcanea canon content (Guardian names, archetypes) — those live in frankxai/arcanea-ecosystem under CC-BY-NC.
  • Theoretical attacks without a working proof-of-concept.
  • Missing-best-practice findings without a concrete exploitation path (e.g., "you don't use [feature X]"). File those as a regular issue or PR with a discussion.

Hall of fame

We credit researchers in our CHANGELOG.md and in the relevant fix commit's Co-Authored-By: trailer (with your consent). If you'd rather stay anonymous, we honor that too.


Built on SIP · SECURITY.md v1.0 · MIT

There aren't any published security advisories