v3.0.0
AgentOps Workbench v3.0.0
Breaking product simplification. Fewer commands and fewer concepts for the
user: two get-data-in intents and three save kinds. No change to ingestion
fidelity, report/gate/export output shapes, adapters, or the dashboard UI.
Migration
Ingest verbs: 4 → 2
There are now two intents instead of four overlapping commands.
| Removed in v3.0.0 | Use instead |
|---|---|
agentops capture codex|claude <prompt> |
agentops run codex|claude <prompt> --no-ingest (write the artifact only) |
agentops capture ... --ingest |
agentops run codex|claude <prompt> (ingests by default) |
agentops import <artifact> |
agentops audit <artifact> --quiet (ingest only, no verdict) |
agentops run keeps all the former capture options (--output, --ephemeral,
--sandbox, --model, --profile, --include-hook-events,
--no-session-persistence, --permission-mode).
Save kinds: 6 → 3
| Removed in v3.0.0 | Use instead |
|---|---|
agentops save repo-json |
agentops save json --repo |
agentops save trace |
agentops save json --format openinference |
agentops save gate |
agentops check --save |
agentops save is now report|pr|json. Removed kinds print a migration hint
instead of failing silently.
Also in this release
- Removed the stale wiki release step from the release template (the GitHub wiki
is disabled).
Verification
bun run ci green (93 tests), smoke:dashboard and smoke:large-session pass.
Source-clone-with-Bun remains the supported install path; npm publication stays
deferred.