Skip to content

OpenClaw test samples

synthetic-* retains original synthetic data with syn- prefixes and constant placeholders. It tests rejection of unknown schemas and legacy migration input. real-2026.9.8 comes from the official npm client on 2026-10-06 in an isolated container with networking disabled: two real local-model calls through public CLI/session continuation, extracted with selected fields and anonymization.

Official documentation (A09, checked 2026-09-24)

Earlier synthetic table names/entry fields were invented placeholders (synthetic-table-* and syn-*). They test discovery shape and rejection, without representing the real schema.

Scenarios

Directory Coverage
synthetic-runtime-store agents/main/agent/openclaw-agent.sqlite with placeholder tables; product identified, table schema unverified, reject
synthetic-legacy-archive legacy agents/main/sessions/ JSONL archives plus sessions.json; migration input remains excluded
real-2026.9.8 Selected exact schema_meta/session_windows/transcript_events fields and independent API usage; uncached input 6,157, output 154, cache read 6,101; default zeros unknown

Expectations are in each directory’s _expectations.md and real-2026.9.8/provenance.json. Native extraction removes prompts, responses, configuration and private paths; stable placeholders replace identities. Rebuilt SQLite contains only required structures. Invalid-type, compression and external-source variants are targeted synthetic cases, without verification of those native product scenarios. Complete originals stay only in root build/. Schema, timestamps and default-zero semantics are in the reading specification.