AI SDK Parity Upgrade

SkillCloud & infra

Upgrade the registered upstream Vercel AI SDK parity baseline for the Go ai-sdk repo. Use when bumping upstream ai or @ai-sdk/* package versions, regenerating conformance expectations, reconciling upstream behavior changes, or moving test/conformance/upstream.yaml forward.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the AI SDK Parity Upgrade skill

What this skill tells your AI

The instructions your AI receives, as published by grafana/ai-sdk in .agents/skills/ai-sdk-parity-upgrade/SKILL.md and read by ahel’s review.

Use this skill to move test/conformance/upstream.yaml to a newer upstream baseline.

Workflow

  1. Read test/conformance/upstream.yaml, test/conformance/PARITY.md, and test/conformance/README.md.
  2. Identify the old upstream target and the affected providers.
  3. Inspect upstream changelog, source, and tests for parity-sensitive changes between the old baseline and the mature stable package set selected from the npm latest release lines. Prefer sources in this order:
    • a local upstream checkout at the registered version or matching tag
    • the conformance tool package installed from the registered baseline
    • raw GitHub source for the registered package version Record which source was used in the PR description when it explains a finding or accepted risk.
  4. Run mise run parity-upgrade to update test/conformance/upstream.yaml, update test/conformance/tools/package.json, refresh test/pnpm-lock.yaml, regenerate conformance expectations, and validate baseline consistency.
  5. Run focused Go tests for implementation areas affected by upstream changes.
  6. Classify every mismatch using the divergence handling table below.
  7. Fix implementation bugs before marking the upgrade complete. Stop and ask before deferring a large design gap that affects public API shape, wire format, provider contracts, or broad harness behavior. Document accepted deviations, gaps, and residual parity risk in the PR description. Do not add broad one-off upgrade findings to test/conformance/upstream.yaml or test/conformance/PARITY.md unless the user explicitly asks or the finding changes a long-lived project policy, coverage map, or baseline contract.
  8. Update the coverage map when the upgrade adds a new provider capability, stream behavior, frontend interop concern, or harness limitation.

Divergence Handling

ClassificationDefault action
Implementation bugAdd or update the relevant fixture when practical, fix the Go implementation, and rerun focused checks.
Upstream behavior changeTrace the change to upstream source or tests, regenerate the matching conformance artifact, adapt Go behavior, and rerun parity checks.
Intentional Go deviationKeep only with a rationale. Document in the PR description unless it changes a durable baseline contract.
Coverage gapAdd a fixture or test when cheap and relevant. Otherwise document the residual risk in the PR description.
Large design gapStop and ask before deferring when the gap affects public API shape, wire format, provider contracts, or broad harness behavior.

Rules

  • Do not update the registered baseline without reviewing generated expected.jsonl and expected-requests.jsonl diffs.
  • Always upgrade to the newest coherent stable package set from the npm latest release lines that satisfies test/pnpm-workspace.yaml's minimumReleaseAge; do not bypass the age gate or pin a beta, canary, or partial package set unless the user explicitly asks for that exception.
  • Do not treat snapshot churn as automatically correct; trace behavior changes to upstream source or tests.
  • Treat regenerated conformance as the upgrade confidence suite: provider implementation changes should be reviewed through request snapshots, and core stream behavior should be reviewed through UI chunk snapshots.
  • Live provider recording remains manual and must not be required for normal CI.
  • Keep package pins, lockfile changes, regenerated snapshots, and baseline metadata in the same change when practical. Keep residual finding summaries in the PR description unless they need durable parity metadata.
  • End the PR description with a short residual-risk section listing accepted gaps, intentional deviations, skipped fixtures, and follow-up owners when known.

Signals

GitHub stars
252
Forks
12
Last commit
Sep 2026
Advanced
Catalog kind
skill
Gateway key
ai-sdk-parity-upgrade
Source
github.com/grafana/ai-sdk