The public package is pre-1.0; the current published baseline is 0.11.0. Security maintenance is focused on the latest 0.11.x release line and future current releases. Older pre-1.0 lines are not promised backports.

Publishing a new package version does not change downstream applications pinned to an exact version; consuming hosts must opt into any upgrade. This table describes the repository's support policy, not a certification or evidence that any particular release is free of vulnerabilities.

Please do not open a public issue with exploit details, credentials, private payloads, or reproduction secrets.

Use GitHub's private vulnerability reporting / security advisory flow for this repository when available. If the repository UI does not expose a private reporting option, open a minimal public issue asking the maintainer for a private contact channel without including vulnerability details.

Useful reports include:

• affected version/commit;
• impacted endpoint, protocol object, or runtime adapter;
• minimum reproduction;
• security impact;
• whether the issue requires authenticated access;
• suggested mitigation, if known.

Event Intelligence 0.11.x is an experimental reference implementation with scoped multi-tenant host support. It is not a turnkey hardened hosted service.

Important defaults and constraints:

• tenant/workspace isolation is implemented through scoped store partitions; tenant-facing code should receive only the corresponding scoped host view;
• protected standalone control-plane endpoints require SERVICE_AUTH_TOKEN;
• GitHub ingress verifies provider HMAC when configured;
• standalone runtime wake packets are HMAC-signed and runtime targets are operator-configured;
• embedded wakes preserve the configured continuation separately from untrusted matched-event evidence;
• event receipt never grants new downstream authorization;
• arbitrary public HTTP wake targets are rejected;
• derived-event lineage is refs-only by default;
• audit records are hash-linked and tamper-evident, not tamper-proof;
• the bundled JSONL store is single-process reference infrastructure and is not safe as a shared multi-replica backend;
• horizontally scaled custom stores must preserve scope isolation and implement wake claim/lease operations atomically;
• secrets belong in environment variables / secret managers, never committed files.

See docs/SECURITY-MODEL.md for the architectural model.