[aurorascharff]

Summary

Production tests need the Instant Navigation testing API enabled during both the build and server startup. With the flag set only during the build, initial page loads render request-time content during instant() tests while client navigations still pause.

The command examples now set the flag for both stages. The skill templates link to the config reference instead of repeating its explanation, while retaining their environment and server lifecycle instructions. The static-shell eval also passes the flag to its Playwright server.

Addresses agent feedback finding 5264.

Updated files

• docs/01-app/03-api-reference/05-config/01-next-config-js/exposeTestingApiInProductionBuild.mdx
• docs/01-app/02-guides/instant-navigation.mdx
• skills/next-cache-components-optimizer/rig-template.md
• skills/next-partial-prefetching-adoption/rig-template.md
• skills/next-partial-prefetching-optimizer/rig-template.md
• evals/evals/agent-050-optimize-static-shell/playwright.config.ts
• evals/evals/agent-050-optimize-static-shell/instant-nav.rig.md

Verification

• Reproduced the original setup with matching Next.js and @next/playwright versions 16.4.2 and 16.5.0-canary.45. Restarting the same build with the variable enabled fixed initial page locking on both versions.
• Passed 12 Playwright tests on 16.5.0-canary.45. Each test checks that request-time content stays absent during the lock and appears after release. Initial page tests also inspect the response markup.
• Tested the exact TypeScript and CommonJS config examples with manual startup. Verified both Playwright startup configurations against a production build, then checked development without an opt-in and a Webpack production build.
• Confirmed that enabling the flag only at startup cannot add the client lock. That configuration also produced a document hydration error.
• Formatting, scoped ESLint, and diff checks passed. The final style edit preserves all code blocks and executable commands.

Full repository lint previously failed on unused Rust task items in untouched files. Process cleanup also reported a sandbox permission error. The full agent-eval suite was not run.

[github-actions]

Tests Passed

Commit: 86c64dd
View workflow run

[vercel]

Vercel Agent reviewed ac3ff4d and found no issues.

[vercel]

Vercel Agent reviewed 4c6f3bd and found no issues.

[vercel]

Vercel Agent reviewed d7de843 and found no issues.