block
Regenerate canonical models when release branch is created.
How this will work:
Runs build_canonical_models -> creates canonical_models.json Runs canonical_model_checker -> creates canonical_mapping_report.json
canonical_model_checker needs secrets to access our common providers.
canonical_mapping_report.json isn't that useful, but what canonical_model_checker prints to stdout about which model mappings were changed/added is, so we might want to expose that a little better.
Can discuss more, but I think on release is a reasonable cadence to update the canonical model mapping + gives us a chance to review if anything catastrophic happened.
Makes sense to me to put this here. How likely do you think it is to fail? If it fails, the release PR never gets created, which is not great but if it's pretty reliable that's good enough.
I don't think these should be "examples" though. You can create regular binary targets with like this:
https://github.com/block/goose/blob/ae8ecbe0906fd972246be62e09b00ee1679c9b0b/crates/goose-server/Cargo.toml#L57
and then run with
cargo run --bin build_canonical_models
Done. Made a new binary and combined these.
Only reasons it would fail are:
- Network issues openrouter/breaking schema change
- Missing secret for the checker
Should both generally be fine and I'd want to know if it breaks.