Establish whether the reported behavior occurs under a controlled, relevant environment. Do not diagnose, verify intent, or fix.
Always update the Reproduction section of tmp/triage/gh-<number>/report.md, including failures and limitations.
Read the Intake section. Record the current commit, branch, package versions, runtime versions, and initial git status --short. Call this the current checkout, not main, unless its commit is verified to match the repository's default branch. If pre-existing modifications touch files that could influence the result — source, tests, configuration, lockfiles, or build outputs — reproduce in an isolated copy of the recorded commit or record the limitation with the outcome.
Use dependencies already installed when possible. If installation is required, use the repository lockfile and a frozen install. For an external or published-version reproduction, work only in the triage directory, inspect manifests first, and disable lifecycle scripts by default.
Do not edit library/playground.ts for source-level reproduction: it imports library/dist/index.mjs, so it tests the last build rather than the current source.
Record:
- Minimal code and steps
- Observed and expected behavior
- Valibot/package, TypeScript, runtime, framework, and OS versions when relevant
- TypeScript flags such as
strictandexactOptionalPropertyTypes - Whether the failure concerns runtime source, type inference, emitted declarations, package exports, a browser, a codemod, or an external integration
Reconstruct a minimal example from the description when faithful. If a linked sandbox or repository is essential, inspect it before running anything. If the expected behavior or key setup cannot be determined, return needs-information and specify the smallest missing facts.
For core library/ and packages/to-json-schema/ source behavior, create a uniquely named temporary test only after confirming the path is absent:
- Runtime:
src/issue-<number>.test.ts - Types:
src/issue-<number>.test-d.ts
Follow nearby tests and import current source through ./index.ts or the package's corresponding relative entry point. Encode the reporter's expected behavior as an assertion, then confirm any failure is the claimed mismatch rather than an import, setup, or unrelated type error.
Run with an explicit package context so the command cannot silently target the wrong package:
pnpm -C library vitest run issue-<number>
pnpm -C library vitest run --typecheck issue-<number>Replace library with the owning package directory, such as packages/to-json-schema, when the report targets another package. Run reproduction commands non-interactively and bound potentially long-running or untrusted reproductions with a timeout instead of letting them hang.
Use the owning package's existing harness for website and codemod reports. For packages/i18n/, test the generated module or build script implicated by the report and run its TypeScript lint check. Do not force every package through the core-library test shape.
Source type tests are insufficient for bugs that appear only across package boundaries, in published exports, or in emitted .d.ts files. For those reports, build or pack the affected package, install it into a minimal consumer under the triage directory with scripts disabled, and reproduce with the reported TypeScript version and tsconfig flags.
For browser-, Deno-, Bun-, or framework-specific behavior, use the named runtime when available. Otherwise return blocked-by-environment; do not claim the bug is absent from a Node-only substitute.
Test the current checkout first. If the issue names an older release, reproduce that exact release in an isolated consumer under the triage directory. Do not conclude already-fixed-on-default-branch unless the default-branch commit was identified and the same reproduction passed against that commit in an isolated clean copy.
If a result differs by version, identify the fixing or regressing commit when practical, but leave causal analysis to the diagnosis phase.
Run at least one nearby control case to prove the harness is sound. Classify the outcome as one of:
reproduced— the precise observed/expected mismatch occursnot-reproduced— the relevant setup was matched and the claimed mismatch does not occurneeds-information— the claim cannot yet be made testablealready-fixed-on-default-branch— reproduced in the reported version but not at the verified default-branch commitblocked-by-environment— a required environment cannot be exercised
Copy the final reproduction into the triage directory. Remove only temporary files created for the run. Compare the final worktree status with the recorded baseline and report any difference; do not demand a globally clean worktree.
Record the exact reproduction code, commands, relevant versions, observed output, expected output, control case, outcome, and confidence in the environment match. Never omit failed attempts that materially limit the conclusion.