Skip to content

Security: KeWang0622/pantheon-skill

Security

SECURITY.md

Security & Privacy Policy

Pantheon handles some of the most sensitive data anyone has: the records, messages, and memories of people they have lost. We take security and privacy extremely seriously.

Threat model

The realistic threats to a Pantheon user are:

  1. Local file exfiltration. Someone gaining access to the user's machine can read ~/.pantheon/souls/*/memory.md and soul.md and learn deeply personal information.
  2. Inadvertent upload. A user accidentally sharing a soul archive (e.g. by adding ~/.pantheon/ to a synced cloud folder, screen-sharing, or pasting a soul file into a chat).
  3. Cross-soul leakage. A bug in family_graph causing one soul to surface another soul's private content inappropriately.
  4. Reconstruction misuse. Someone using a Pantheon-built soul to impersonate the deceased in a way that harms surviving family members.

We do not currently treat passive network attackers as in-scope because Pantheon does not transmit data over the network in its standard configuration. If that ever changes, this section will be updated and the change called out in CHANGELOG.md.

What Pantheon does to mitigate

  • All data stays local. No telemetry. No analytics. No "anonymous usage data." Pantheon imports the standard library, pillow (optional, for photo EXIF), and pytest (dev only). No outbound network calls in normal operation.
  • ~/.pantheon/ is the entire storage footprint. Nothing is written outside it except temporary scratch in /tmp when explicitly invoked.
  • Soul archives are human-readable Markdown + JSON. No proprietary binary formats. Users can audit, edit, or delete any archive with a text editor.
  • Versioning + rollback. Every update creates a snapshot via tools/version_manager.py. Mistakes are reversible.
  • The /pantheon-demo flow never overwrites real archives — it falls back to a demo_ prefix on slug conflict.
  • is_example: true and fictional: true flags in demo souls' meta.json force the runtime to display "This is a fictional example" at the top of every dialog.

What Pantheon explicitly does NOT do

  • Pantheon does not upload soul archives anywhere. There is no cloud service, no shared registry, no "social" features.
  • Pantheon does not use your soul data to train any model.
  • Pantheon does not send analytics or telemetry of any kind.
  • Pantheon does not embed advertising or any third-party tracker.

Reporting a vulnerability

If you discover a security issue — anything from a path-traversal bug in a parser to a leak between souls in the family graph — please report it privately rather than opening a public issue.

Preferred channel: GitHub private vulnerability reporting → Report a vulnerability.

Email: the address listed in package.json (when present) or DM the maintainer on GitHub.

Please include:

  • A clear description of the vulnerability.
  • Steps to reproduce against the /pantheon-demo Wang family example wherever possible (so we never need to look at your real data).
  • Any proof-of-concept code (sanitized).
  • Your suggested fix, if any.

We will respond within 7 days with an acknowledgment and within 30 days with a fix or a written rationale if a fix is not feasible. We will credit you in the security advisory unless you ask us not to.

What Pantheon will not pay for

Pantheon does not currently run a bug bounty program. We are an open-source project run by individual maintainers. If a vulnerability you report leads to a coordinated disclosure, we will credit you, thank you publicly, and remember you fondly.

If you are a journalist or researcher

You are welcome to study Pantheon. Please use the /pantheon-demo Wang family example for reproduction — those souls are fictional composites and we maintain them precisely so researchers, journalists, and security testers never have to work against real grief data.

There aren't any published security advisories