• THREAT_MODEL.md — assets, trust boundaries, threat actors, mitigations
• PRIVACY.md — what data is handled and where it goes
• COMPLIANCE.md — GDPR / CCPA / HIPAA / SOC 2 / PCI / ISO posture
• docs/SECURITY_REPORT.md — full audit-style assessment with 13 numbered findings
• docs/SECURE_DEPLOYMENT.md — operator hardening guide (network, container, persistence, monitoring)
• docs/ARCHITECTURE.md — system architecture with Mermaid diagrams (sections 5-6 cover trust boundaries and threat surface)
• legal/DPA-template.md — Data Processing Addendum template

If you discover a security vulnerability in snapfeed, please do not open a public GitHub issue.

Instead, email shimoverse@gmail.com with:

• A description of the vulnerability
• Steps to reproduce
• The version affected (npm list snapfeed or commit SHA)
• Your assessment of the impact

You should expect an acknowledgement within 3 business days and a fix plan within 10 business days for confirmed issues.

We follow coordinated disclosure: please give us a reasonable window to ship a fix before publishing details.

We patch security issues on the latest minor version. While snapfeed is pre-1.0, we do not backport to earlier versions.

If your security team needs to approve snapfeed before adoption, here is what they will likely ask. Tick each item against the current release.

• Zero phone-home. snapfeed makes no telemetry, analytics, or "update check" calls to any domain we control. The only outbound calls are to adapters you explicitly configure.
• Outbound allowlist. Only the destinations you wire up (Slack webhook, JIRA URL, your Postgres, etc.) receive data.
• Self-hostable. All server code can run inside your VPC. No required SaaS dependency.
• Open source, MIT. Audit the full source tree. No CLA, no obligation.

• LLM is optional. snapfeed works fully without any GenAI calls. Every smart feature degrades gracefully (see "LLM degradation table" in the README).
• Bring your own key. When LLM is enabled, you supply the API key and provider. snapfeed never proxies through our servers.
• In-tenant LLMs supported. Compatible with Azure OpenAI in your tenant, AWS Bedrock in your account, and self-hosted Ollama / vLLM. No data leaves your boundary.
• Server-side only. API keys are never exposed to the browser bundle.
• Pre-LLM redaction. Optional regex pass strips emails, tokens, and high-entropy strings before any LLM call.

• Secrets stay server-side. Adapter tokens (Slack webhook, GitHub PAT, Supabase service key) are read from process.env on the server and are never sent to the client.
• Console error redaction. A built-in regex pass strips token=, key=, secret=, Authorization, and JWT-shaped strings from captured console errors before they leave the browser.
• Payload size caps. 10 KB text + 5 MB screenshot by default; configurable.
• Origin allowlist. Server handler rejects requests from non-allowlisted origins when configured.
• Rate limiting. In-memory by default; Redis/Upstash supported for distributed deployments.

• Disabled in production by default. Widget is a no-op unless enableInProduction: true is set explicitly.
• Role-scopable. When enabled in production, you can scope by user role: enableInProduction={user.role === 'admin'}.

• Pinned dependencies. package-lock.json checked in.
• Minimal runtime dependencies. Zero hard runtime deps; html2canvas is an optional peer.
• No eval, Function(), or dynamic remote imports in the bundle.
• Audit log primitive shipped in v0.4 — snapfeed/audit-log exposes fileAuditLog, noopAuditLog, multiAuditLog with a discriminated AuditEvent union for feedback.received, adapter.dispatched, llm.called, config.changed, rate_limit.hit. Wire into FeedbackHandlerConfig.onReceive / onComplete.
• Self-hostable Docker stack shipped in v0.4 — docker/docker-compose.yml runs the worker + MinIO (+ optional Ollama for in-tenant LLM). No outbound calls during build or runtime beyond what you configure.
• SBOM published per release. Slipped to v0.6 (npm sbom + GitHub Actions artifact).
• Reproducible builds. Partially: package-lock.json pinned + Dockerfile builds deterministically. Image digests still TODO for v0.6.

• Retention policy + GDPR right-to-erasure. Time-based retention (pruneOlderThan({ retentionDays: N })) shipped in v0.6. User-initiated deletion (deleteByUserId(reporter, { auditLog, storage })) shipped in v0.7 — see docs/gdpr.md.
• SSO/SAML for admin. OIDC + SAML for the self-hosted admin app. Slipped to v0.6.
• Image digest pinning in docker/docker-compose.yml. Shipped in v0.6 — compose file pins to specific tags (no :latest) and ./docker/pin-digests.sh resolves + applies sha256 digests for supply-chain-grade reproducibility.

• We do not run a hosted SaaS. You install snapfeed in your own infrastructure.
• We do not collect any telemetry from installs.
• We do not require you to sign a CLA to contribute.
• We do not have a "free tier" or paid tier — the library is MIT, period.

If a future release ever changes any of the above, it will be called out in the changelog and require a major version bump.