Files
learnharness/tests/byok-refusal.test.ts
T
amosblomqvist e4049e4bf8 Key-presence refusal at every gate: a pinned model is not enough
Owner-diagnosed: the turn gate checked only model coordinates, so a
pinned slot whose provider key was deleted passed the gate and died
mid-stream masked to 'An error occurred.' The key is now gated where
the model is — by what would RUN (env-first keyValue truth, so an
env-shadowed key passes).

- core/settings.ts: the provider_key_missing envelope family (code,
  provider, slots, human line naming the provider from AUTH_PROVIDER_KEYS
  labels) + ProviderKeyMissingError + isProviderKeyRefusal matcher.
- chat-handler turn gate: refuses 503 before beginTurn when the resolved
  conversation's provider key is gone (slots: [strong]).
- Job runner: same check via an injected dep (agent layer cannot read
  the key store) — a keyless engine refuses the spawn through the
  existing failure-shape convention, no phantom run.
- Extraction pipeline: refuseIfStagesUnready refuses BEFORE any batch
  (first offender: extractor before auditor); the trigger skips info-
  level, /api/graph/import surfaces the 503 envelope per file.
- Tests: 8 new pins (turn gate both ways, runner refusal, pipeline
  first-offender + pass-through, envelope factory, matcher); the three
  turn-driving suites seed OPENROUTER_API_KEY like any real setup.
- Doc fix: stale EXTRACTION.chunkTokens comment in the manual-extract
  route (left over from eed4b4e).

Known residual (pre-existing, flagged for master): the manual extract
route logs BOTH config refusals as console.error (fire-and-forget path);
the coverage ledger keeps everything uncovered either way.
2026-09-24 22:51:12 +03:00

152 lines
6.7 KiB
TypeScript

import { describe, it } from "node:test";
import assert from "node:assert/strict";
import { parseModelsRefusal } from "../components/chat.tsx";
import { prefillFromSlots, prefillLevelFromPin, setupStage } from "../components/setup-dialog.tsx";
import {
isModelsRefusal,
isProviderKeyRefusal,
modelsNotConfigured,
providerKeyMissing,
PROVIDER_KEY_MISSING_CODE,
} from "../src/core/settings.ts";
import type { ModelCatalogEntry, ModelSlotName } from "../src/core/index.ts";
/**
* The BYOK universal gate rendered honestly (owner, 2026-09-22): the server
* refuses ALL turns with 503 {error, code: "models_not_configured", missing}
* when a slot tier is unset. These pin the client's pure seams: the refusal
* detector (the transport throws the 503 body verbatim as the error text),
* the job-record refusal matcher, and the SetupDialog's prefill.
*/
const REFUSAL = JSON.stringify(modelsNotConfigured(["strong"]));
describe("parseModelsRefusal — the send refusal detector", () => {
it("reads the missing tiers from the envelope text", () => {
assert.deepEqual(parseModelsRefusal(REFUSAL), ["strong"]);
assert.deepEqual(parseModelsRefusal(JSON.stringify(modelsNotConfigured(["strong", "vision"]))), [
"strong",
"vision",
]);
});
it("anything else is a real error — null, the honest banner stays", () => {
assert.equal(parseModelsRefusal(undefined), null);
assert.equal(parseModelsRefusal("Something went wrong"), null); // not JSON
assert.equal(parseModelsRefusal('{"error":"A turn is already running"}'), null); // JSON, wrong code
assert.equal(parseModelsRefusal('{"code":"models_not_configured","missing":"strong"}'), null); // missing not an array
assert.equal(parseModelsRefusal('{"code":"models_not_configured","missing":[42]}'), null); // tiers are strings
});
});
describe("isModelsRefusal — the job-record matcher", () => {
it("matches the runner's composed wording, not real failures", () => {
assert.equal(isModelsRefusal(`${modelsNotConfigured(["vision"]).error} (missing: vision)`), true);
assert.equal(isModelsRefusal(modelsNotConfigured(["cheap"]).error), true);
assert.equal(isModelsRefusal(undefined), false);
assert.equal(isModelsRefusal("FAILED: headless crashed"), false);
});
});
describe("providerKeyMissing — the envelope factory (provider-key interlude, 2026-09-25)", () => {
it("names the provider from the curated label and carries the exact machine code", () => {
const envelope = providerKeyMissing("openrouter", ["strong"]);
assert.deepEqual(envelope, {
error: "No OpenRouter key is set — add one in Settings → Keys",
code: PROVIDER_KEY_MISSING_CODE,
provider: "openrouter",
slots: ["strong"],
});
assert.equal(envelope.code, "provider_key_missing");
});
it("an unknown provider falls back to its raw id as the label", () => {
const envelope = providerKeyMissing("weird", ["cheap"]);
assert.equal(envelope.error, "No weird key is set — add one in Settings → Keys");
});
});
describe("isProviderKeyRefusal — the job-record matcher (provider-key interlude)", () => {
it("matches the envelope wording regardless of provider label, not real failures", () => {
assert.equal(isProviderKeyRefusal(providerKeyMissing("together", ["cheap"]).error), true);
assert.equal(isProviderKeyRefusal(providerKeyMissing("openai", ["strong"]).error), true);
assert.equal(isProviderKeyRefusal(""), false);
assert.equal(isProviderKeyRefusal(undefined), false);
assert.equal(isProviderKeyRefusal("FAILED: headless crashed"), false);
assert.equal(isProviderKeyRefusal(modelsNotConfigured(["strong"]).error), false, "the models refusal is a different failure");
});
});
describe("prefillFromSlots — the SetupDialog prefill (BYOK wave)", () => {
const entry = (id: string, provider: string): ModelCatalogEntry => ({
id,
name: id,
provider,
reasoning: false,
input: ["text"],
supportedLevels: ["low"],
cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
contextWindow: 8192,
maxTokens: 1024,
});
const catalog = [entry("glm-5.3-flash", "z-ai"), entry("qwen3.8-flash", "qwen")];
const pins = (strong: boolean, vision: boolean) => ({
strong: strong ? { model: { provider: "z-ai", id: "glm-5.3-flash" } } : null,
cheap: null,
vision: vision ? { model: { provider: "qwen", id: "qwen3.8-flash" } } : null,
});
it("prefills exactly the pinned tiers, matched by provider AND id", () => {
const pre = prefillFromSlots(catalog, pins(true, true));
assert.equal(pre.strong?.id, "glm-5.3-flash");
assert.equal(pre.vision?.id, "qwen3.8-flash");
assert.equal(pre.cheap, undefined);
});
it("nothing pinned (first-run) → nothing prefilled — all three marked needed", () => {
const pre = prefillFromSlots(catalog, pins(false, false));
assert.deepEqual(pre, {});
});
it("a pin whose model left the federated catalog does not prefill (lapsed key)", () => {
const pre = prefillFromSlots(catalog, { strong: { model: { provider: "gone", id: "vanished" } }, cheap: null, vision: null });
assert.deepEqual(pre, {});
});
it("provider collisions never cross-pollute (same id, different provider)", () => {
const both = [entry("shared-id", "openai"), entry("shared-id", "openrouter")];
const pre = prefillFromSlots(both, { strong: { model: { provider: "openrouter", id: "shared-id" } }, cheap: null, vision: null });
assert.equal(pre.strong?.provider, "openrouter");
});
it("absent slots payload (older shape) prefills nothing", () => {
const pre = prefillFromSlots(catalog, undefined);
assert.deepEqual(pre, {});
});
});
describe("prefillLevelFromPin — the setup dialog's level prefill (chunk 2)", () => {
const entry = (levels: string[]) =>
({ id: "m", name: "M", provider: "p", reasoning: true, input: ["text"], supportedLevels: levels, cost: { input: 1, output: 1, cacheRead: 0, cacheWrite: 0 }, contextWindow: 1000, maxTokens: 1000 }) as never;
it("a stored level the pinned model still supports prefills as-is", () => {
assert.equal(prefillLevelFromPin(entry(["low", "high"]), { thinkingLevel: "high" }), "high");
});
it("a stored level the model LOST rides as-is — Save clamps server-side (bd68486)", () => {
assert.equal(prefillLevelFromPin(entry(["minimal", "medium"]), { thinkingLevel: "xhigh" }), "xhigh");
});
it("a legacy pin with no stored level prefills the ordering's level", () => {
assert.equal(prefillLevelFromPin(entry(["low", "medium"]), {}), "medium");
assert.equal(prefillLevelFromPin(entry(["low", "medium"]), null), "medium");
});
});
describe("setupStage — unchanged by the BYOK wave", () => {
it("keys first (owner ordering), pickers when a provider is keyed", () => {
assert.equal(setupStage(null), "keys");
assert.equal(setupStage([{ id: "x", keyPresent: false }]), "keys");
assert.equal(setupStage([{ id: "x", keyPresent: true }]), "pickers");
});
});