Context
Integration tests were removed from CI in #36 to unblock plugin-loading isolation work. They run locally only today.
What needs to happen
- Decide which integration scenarios are worth running on every PR vs only on
workflow_dispatch.
- Re-add the
integration job to .github/workflows/ci.yaml with if: github.event_name != 'push' so it runs on PRs and manual dispatch but not on every push to main.
- Re-add
Integration tests to the required-status-check contexts in .github/settings.yml for main.
- Confirm the chosen test model (currently
opencode/minimax-m2.5 paid by default; OPENCODE_TEST_MODEL overrides) does not blow through Copilot quota under PR load.
Why this is parked
The free models (opencode/big-pickle, opencode/minimax-m2.5-free) hit https://opencode.ai/zen rate limits (HTTP 429) under repeated runs and silently retry until tests time out. Local runs use a paid model; that 's not appropriate to default in CI without first scoping cost.
Done when
- Integration job runs on PRs and manual dispatch.
- Required status check is back on
main.
- A short note in
AGENTS.md documents the model-override env var and the cost characteristic.
Context
Integration tests were removed from CI in #36 to unblock plugin-loading isolation work. They run locally only today.
What needs to happen
workflow_dispatch.integrationjob to.github/workflows/ci.yamlwithif: github.event_name != 'push'so it runs on PRs and manual dispatch but not on every push tomain.Integration teststo the required-status-checkcontextsin.github/settings.ymlformain.opencode/minimax-m2.5paid by default;OPENCODE_TEST_MODELoverrides) does not blow through Copilot quota under PR load.Why this is parked
The free models (
opencode/big-pickle,opencode/minimax-m2.5-free) hithttps://opencode.ai/zenrate limits (HTTP 429) under repeated runs and silently retry until tests time out. Local runs use a paid model; that 's not appropriate to default in CI without first scoping cost.Done when
main.AGENTS.mddocuments the model-override env var and the cost characteristic.