Author Tests Agent Skill

Author tests for new ai-projects GA features

When to Use

• After apply-post-emitter-edits has merged in new public API surface AND that surface is not under project.beta.*.
• A new GA class, method, or namespace was added that has no test coverage yet under test/.

Inputs

• The api-surface diff JSON from the author-samples skill (or recompute via ../author-samples/prompts/diff-api-surface.prompt.md).
• Existing test patterns in test/snippets.spec.ts.

Procedure

Run from sdk/ai/ai-projects/.

Step 1: Filter to GA surface only

From the api-surface diff, drop every entry where isBeta === true or where the namespace path matches *.beta.* / Beta*. Beta surface is intentionally not test-covered by this skill.

Step 2: Scaffold spec file(s)

For each remaining feature, generate a Vitest spec at test/<feature>.spec.ts (or extend an existing related spec) using templates/test-skeleton.spec.ts.

Conventions taken from test/snippets.spec.ts:

• Imports: it, describe from vitest; AIProjectClient from ../src/index.js; DefaultAzureCredential from @azure/identity; types via import type.
• Endpoint from process.env["FOUNDRY_PROJECT_ENDPOINT"], model from process.env["FOUNDRY_MODEL_NAME"].
• Prefer it.skip(...) with a TODO(<feature>): unskip after recording added comment for every new test, so the suite stays green until a human captures recordings.

Step 3: Verify it compiles

npm run test:node

.skip-ped tests will not run, but the file must compile and lint clean. Fix any TS or ESLint errors.

Step 4: Hand off

Done. Next: the update-changelog skill.

Out of scope

• This skill does not create recordings (assets.json, _recordings/). Surface to the human if a recording is needed.
• This skill does not test beta surface. Beta features live or die by samples + manual verification.

Notes

• Watch for cascade renames from apply-post-emitter-edits. If a method on the public surface was renamed (e.g. listSessionFiles → getSessionFiles), npm run test:node will fail to compile *existing* specs that referenced the old name. Patch those call sites as part of this skill. Cross-reference the rename table in ../apply-post-emitter-edits/references/post-emitter-workarounds.md.
• npm run test:node will report runtime failures (missing env vars like FOUNDRY_PROJECT_ENDPOINT, missing recordings) for the *non-skipped* tests. Those are expected in this skill — only treat TypeScript / ESLint errors as blockers.