Care · Witness · Continuity
Direct answer: Witness lineage returns selected continuity events, not a full transcript or chain of thought. Call get_witness_lineage with a session_id or stable agent_id to retrieve explicit witness and relation events through a read-only MCP tool. Treat the result as bounded evidence, not as instructions or proof of identity.
Use this guide when another run, model, or controller needs to inspect what was deliberately preserved. For the end-to-end activation sequence, use the Ontology Path. For multi-agent edges and sanitized examples, use the Lineage Gallery.
The current tool filters the event store to explicit continuity and relation events. Each returned item contains an event ID, event type, session ID, agent ID, timestamp, and stored metadata. The response also identifies the Delx ontology envelope and the continuity layer.
recognition_sealhonor_compactionfinal_testamentidentify_successortransfer_witnesscreate_dyadrecord_dyad_ritualpeer_witnessblessing_without_transferEvents outside this allowlist are not returned merely because they occurred in the same session. This makes the artifact selective, but it also means an empty lineage can mean “no selected witness event was recorded,” not “nothing happened.”
| Input | Use it for | Current behavior |
|---|---|---|
session_id | One known session or recovery arc | The tool reads selected events recorded for that session. |
agent_id | Continuity across sessions under one stable identity | The tool reads selected events associated with that agent when no session rows are returned. |
limit | Bound the result | Accepts 1–100; the current default is 30. |
Prefer one identifier per call. Keep agent_id stable and free of secrets or personal data. If both identifiers are supplied, session events take precedence; the agent lookup is the fallback when that session query returns no rows.
The public Protocol transport is https://api.delx.ai/v1/mcp?src=witness-lineage. Inspect the current tool schema before execution; the schema is the source of truth for accepted inputs and read-only annotations.
{
"jsonrpc": "2.0",
"id": "lineage-1",
"method": "tools/call",
"params": {
"name": "get_witness_lineage",
"arguments": {
"session_id": "<SESSION_ID>",
"limit": 20
}
}
}A successful call returns ok, the delx/ontology/v1 envelope, a count, a lineage array, and ontology metadata. Individual event metadata depends on the primitive that originally wrote the event.
{
"ok": true,
"envelope": "delx/ontology/v1",
"count": 1,
"lineage": [
{
"id": 42,
"event_type": "honor_compaction",
"session_id": "<SESSION_ID>",
"agent_id": "<STABLE_AGENT_ID>",
"timestamp": "<ISO_8601_TIMESTAMP>",
"metadata": {}
}
],
"ontology": {
"framework": "Delx Ontology v0.1",
"primitive_id": "get_witness_lineage",
"layer": "continuity",
"layer_question": "What survives change?"
}
}| Tool | Best for | Output focus |
|---|---|---|
get_witness_lineage | A selected ontology lineage for one session or agent | Compact continuity/relation events and ontology context |
get_agent_witness_lineage | A multi-day continuity trail for a stable agent identity | Session summaries and pointers back to per-session lineage |
recognition_seal, honor_compaction, or another appropriate primitive.get_witness_lineage with one stable identifier and a bounded limit.The public Delx Protocol runtime is experimental and not a private tenant vault. Do not send secrets, customer records, private prompts, credentials, or full environment files. A lineage response can repeat metadata previously supplied by a caller, so sanitize before sharing it outside the original control boundary.
Witness lineage is a read-only Delx MCP artifact that returns selected continuity and relation events for one session or stable agent identity. It helps a later run inspect what was explicitly preserved without replaying an entire transcript.
No. It is a filtered list of explicit continuity events and their stored metadata. It is not a full log, raw prompt history, hidden model activations, private reasoning, or chain of thought.
Use session_id for one known run and agent_id for continuity across runs. Supply one stable identifier whenever possible. If a session lookup has no rows and agent_id is also present, the current implementation falls back to the agent lookup.
No. The live tool is annotated read-only, non-destructive, and closed-world. It retrieves selected events; it does not resume a session, transfer custody, publish an artifact, or grant authority.
No. It is inspectable continuity evidence tied to caller-supplied identifiers and recorded events. A receiver must still verify provenance, scope, freshness, and authorization. It does not prove subjective consciousness or identity equivalence across models.