Every Harbor invocation in our docs pins harbor==0.15.0:
README.md:470, :492, :507, :834
src/README.md:29
- (and the adapter docs generated from these)
Upstream is now 0.21.0 (harbor-framework/harbor, PyPI harbor). Between those versions the dataset interface changed: 0.15's local-path form harbor run -p ./harbor-datasets/clawbench-v2 has given way to hub-published datasets addressed as harbor run -d org/dataset@version, and the hub publish flow (harbor auth login → harbor dataset init → harbor add → harbor publish --public) is now the documented path.
Impact:
- Users following our docs install a six-versions-stale Harbor; anyone who already has current Harbor finds our commands don't match their CLI.
- It blocks the more valuable thing: publishing ClawBench to the Harbor hub as
tiger-ai-lab/clawbench, which makes the benchmark runnable by anyone with harbor run -d tiger-ai-lab/clawbench@v1 and gets hub-side leaderboard/analytics for free. That publish flow requires ≥0.16-era CLI.
Ask:
- Smoke-test
clawbench-harbor-adapt output against Harbor 0.21 and note the supported range in the docs (or bump the pin).
- Decide whether to publish the V2 dataset to the Harbor hub (separate issue if yes).
Every Harbor invocation in our docs pins
harbor==0.15.0:README.md:470,:492,:507,:834src/README.md:29Upstream is now 0.21.0 (harbor-framework/harbor, PyPI
harbor). Between those versions the dataset interface changed: 0.15's local-path formharbor run -p ./harbor-datasets/clawbench-v2has given way to hub-published datasets addressed asharbor run -d org/dataset@version, and the hub publish flow (harbor auth login→harbor dataset init→harbor add→harbor publish --public) is now the documented path.Impact:
tiger-ai-lab/clawbench, which makes the benchmark runnable by anyone withharbor run -d tiger-ai-lab/clawbench@v1and gets hub-side leaderboard/analytics for free. That publish flow requires ≥0.16-era CLI.Ask:
clawbench-harbor-adaptoutput against Harbor 0.21 and note the supported range in the docs (or bump the pin).