test: main-suite hygiene — env-dependent failures skip cleanly, broken ones fixed (#1532)

[aarontrowbridge]

Closes #1532 — the 10 amico-run + 16 extension pre-existing suite failures on main @ 244e6a9, each diagnosed and made honest. No test deleted, no assertion weakened, no test marked broken.

Classification

#	Test (suite)	Verdict	Diagnosis
1	cloud_verb › status › unreachable endpoint degrades to ONE clear line (amico-run)	fixed (test)	The assertion pinned node's fetch failed string; bun's fetch says Unable to connect. Is the computer able to access the url?. Assert the union of both runners' wording — the intent (ONE clear named connectivity failure) is unchanged.
2	cloud_verb › abort › preflights the channel (amico-run)	fixed (test)	Same runner-dependent wording as #1.
3–5	sota_papers › curlSotaFetch ×3 (amico-run)	fixed (product code)	execFileSync("curl", …) relied on default env inheritance; bun resolves unqualified executables from the spawn's env OPTION when given, else the process-START env snapshot — never live process.env mutations — so the tests' fake-curl PATH injection was invisible and the REAL arXiv fetch ran. Fix: pass env: process.env explicitly (byte-identical inheritance under node; live PATH honored under bun).
6	sota_papers › lens 404 round via the production transport (amico-run)	fixed (product code)	Same root cause as #3–5, via the default transport.
7	sota_codebase › loadRegistry bootstraps from the packaged seed (amico-run)	fixed (product code)	The #820 data-as-import contract is TEXT (esbuild .toml: "text", vitest tomlAsText plugin, src/sota_seed.d.ts); bun natively parses .toml imports into an object, so seedToml.endsWith crashed. Normalized back to text at the import boundary (smol-toml stringify; the validator re-parses it either way).
8	doctor › parseDoctorArgs usage error (amico-run)	fixed (test)	A regex nested inside toMatchObject substring-matches under vitest but is compared literally under bun. Message now asserted with toMatch — same strength, both runners.
9	spec_review › the no-real-model-calls guard (amico-run)	skip (env)	Requires the vitest setup file test/setup.ts (it pins $AMICO_CRITIC_BIN to an impossible path for the suite); bun test does not load vitest setup files. it.skipIf(!process.env.AMICO_CRITIC_BIN) — run pnpm test to exercise. The fail-closed behavior itself stays covered by the sibling opt-in case.
10	gh_cli › gh shim recursion guard (amico-run)	fixed (test) + env skip	The constructed PATH assumed dirname(process.execPath) holds node — true under the node runner, false under bun (bun's bin dir has no node; the bundle could not even spawn, exit −1). The node dir is now RESOLVED (realpath'd, so shim dirs that carry a real gh cannot defeat the guard). A host with no node binary at all skips with the requirement named.
11–22	amicode_service_{bootstrap_auth ×7, contract, engine_proxy, fleet_data_plane, wiring ×2} (extension)	fixed (tests)	The dev hosts run a LIVE amicode service whose ambient env exports AMICODE_SERVICE_AUTH=open (the #955 tunnel/LAN posture). Tests boot the service relying on the CREDENTIAL default and never isolated the env → open mode → every anonymous-401 pin returned 200. Save/clear/restore AMICODE_SERVICE_AUTH around those boots (the auth_mode file's env-carrier cases already used the idiom; the boot-log case was missing it — fixed to the same idiom).
23	amicode_service_auth_mode › authMode exposed for the wiring's log line (extension)	fixed (test)	Same leak; this case had forgotten the save/delete/restore its siblings use.
24	amicode_service_runner › a shelf without a built app dist fails with the named reason (extension)	fixed (test)	Pointed at the vendored engine binary, absent on hosts without the vendor fetch → the runner failed EARLIER with no engine binary — the wrong named reason. Now passes an existing stub bin so the DIST check is what fires; the fail-loud describe is truly "always on" as its header claims.
25	cli_gate › the freshly staged CLI passes every check (extension)	skip guard tightened	The guard checked bare bin/ existence; a PARTIAL staging (leftover dist/, no launchers) passed the guard and red-failed the gate. Now requires every declared bin's launcher + dist present — skip names the requirement (pnpm -r run build); with a real staging the test runs and passes.
26–27	terminal › OPENCODE_DB/OPENCODE_CONFIG_DIR injection ×2 (extension)	fixed (tests)	The terminal env is seeded from process.env; the dev host exports OPENCODE_DB, leaking into the no-injection pins. The describe now saves/clears/restores both vars so the settings, not the host env, are what's tested.
28	open_threads › returns null when bun:sqlite is unavailable (extension; bun test surface)	skip (env)	Requires the Node runtime; under bun bun:sqlite exists and the call would read the developer's REAL session DB. describe.skipIf(typeof Bun !== "undefined") — run pnpm test to exercise.

Verification (from package dirs only, fresh state)

packages/amico-run — bun test (the campaign's baseline surface, which reproduced the 10):

 1707 pass
 10 skip
 0 fail
Ran 1717 tests across 102 files.

(baseline on main @ 244e6a9: 1698 pass / 9 skip / 10 fail — the identical 10, byte-for-byte the #1528 record)

packages/amico-run — pnpm vitest run --passWithNoTests --exclude '**/slow/**':

Tests  1708 passed | 2 skipped (1710)

packages/extension — pnpm vitest run --passWithNoTests:

Tests  3680 passed | 61 skipped (3741)

(baseline: 16 failed | 3663 passed | 62 skipped (3741) — the identical 16, byte-for-byte the #1526 record; +1 pass/−1 skip is the cli_gate real-set test running for real after staging)

bun test test/open_threads.test.ts in packages/extension: 47 pass / 1 skip / 0 fail (the Node-runtime degradation case skips under bun with its requirement named; 48/48 under vitest).

bun typecheck in both packages: 0 errors.

Every skip carries its stated requirement in-code (endpoint/credential/binary/runtime + how to run it); skip guards are deterministic per environment.

[coderabbitai]

Important

• 🔍 Trigger review

This repository does not receive automatic reviews because it has fewer than 10 stars.

⚙️ Run configuration

Configuration used: defaults

Review profile: CHILL

Plan: Advanced

Run ID: ebbbbdd3-436f-4d0c-8854-bab5a20c15b0

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands.}