Skip to content

fix(build): Catch harness build failures before merge#43

Merged
dcramer merged 5 commits into
mainfrom
codex/fix-ci-build-gates
May 3, 2026
Merged

fix(build): Catch harness build failures before merge#43
dcramer merged 5 commits into
mainfrom
codex/fix-ci-build-gates

Conversation

@dcramer
Copy link
Copy Markdown
Member

@dcramer dcramer commented May 3, 2026

Restore the harness declaration build and add the missing merge gates that would have caught it. Pull request CI now runs typecheck and the workspace build after lint/tests, so DTS regressions fail before publish-time workflows.

Harness Build Fixes

DTS generation now uses stricter public types for Pi harness result normalization and tool-call errors. AI SDK runtime tool calls without ids are preserved instead of being compared against SDK step ids.

CI Gates

The PR workflow runs release config validation, lint, typecheck, CI tests, and the workspace build before merge. The root test scripts also cover apps, so app-owned demo fixture tests run in CI.

Demo Fixtures

The refund demo runtime is no longer a workspace package under packages/ or examples/. The Pi demo owns its src/refundAgent.ts fixture, and the AI SDK demo owns its local tools/assertion helper. README and package docs keep factuality/LLM-as-a-judge examples as the primary documentation pattern.

Validated locally with pnpm install --frozen-lockfile, pnpm lint, pnpm typecheck, pnpm test:ci, and pnpm run build.

dcramer and others added 3 commits May 3, 2026 10:04
Keep harness adapter types valid under declaration generation and typecheck. Preserve optional runtime tool ids, keep Pi result normalization on the declared result type, and serialize tool-call errors with required messages.

Co-Authored-By: OpenAI Codex <codex@openai.com>
Add typecheck and workspace build to the pull request workflow so declaration and type regressions fail before post-merge publishing jobs.

Co-Authored-By: OpenAI Codex <codex@openai.com>
Keep packages reserved for publishable surfaces by moving the private Foobar demo runtime into examples. Update workspace paths, tests, docs, and README examples to keep factuality judges as the primary documented pattern.

Co-Authored-By: OpenAI Codex <codex@openai.com>
@dcramer dcramer marked this pull request as ready for review May 3, 2026 17:08
dcramer and others added 2 commits May 3, 2026 10:10
Rename the private demo runtime from foobar to refund-agent so the folder and workspace package describe their purpose. Update demo imports, docs, and path aliases to use @demo/refund-agent.

Co-Authored-By: OpenAI Codex <codex@openai.com>
Remove the shared @demo/refund-agent workspace package and keep the refund fixtures inside the demo apps that use them.

The Pi demo now owns its Pi agent directly, while the AI SDK demo owns its simple tool data and assertion helper. Root test scripts cover apps so the app-local fixture test still runs in CI.

Co-Authored-By: OpenAI Codex <codex@openai.com>
@dcramer dcramer marked this pull request as draft May 3, 2026 17:18
@dcramer dcramer marked this pull request as ready for review May 3, 2026 17:23
@dcramer dcramer merged commit 45588a5 into main May 3, 2026
9 checks passed
@dcramer dcramer deleted the codex/fix-ci-build-gates branch May 3, 2026 17:24
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant