fix: keep research available without search keys

This commit is contained in:
amosblomqvist
2026-10-24 06:52:36 +03:00
parent 6cba9c8c5a
commit 4bd20be529
11 changed files with 277 additions and 129 deletions
+124 -31
View File
@@ -1,20 +1,14 @@
/**
* Search rewiring (search port wave, 2026-09-20) — the web_search tool's
* port injection + the research dispatch tool's registry-level drop.
*
* Google CSE is dead: the tool runs on the injected search port
* (core/search.ts), the runner threads the seam (searchProvider — covered
* end-to-end in jobs.test.ts), and jobTools drops the research tool when
* ctx.searchAvailable reports no live engine (owner decision: registry
* drop, no doomed dispatch, no silent fallback).
*
/** Keyless research availability + the injected search port.
* No network, no shared data files — the jobStore fake below is in-memory.
* A missing search key is a recoverable tool result, not a dispatch gate.
*/
import assert from "node:assert/strict";
import { describe, it } from "node:test";
import { createWebTools } from "../src/agent/tools/web.ts";
import { jobTools } from "../src/agent/tools/jobs.ts";
import { runResearcherAgent } from "../src/agent/subagents/researcher.ts";
import { createFauxCore, fauxAssistantMessage, fauxText, fauxToolCall, type FauxResponseStep } from "@earendil-works/pi-ai";
import type { JobStore } from "../src/core/ports.ts";
import type { JobRecord } from "../src/core/contracts.ts";
import type { SearchHit, SearchProvider } from "../src/core/search.ts";
@@ -31,9 +25,13 @@ function stubProvider(hits: SearchHit[] = [{ title: "t", url: "https://e.com/",
describe("web_search on the injected port (web.ts)", () => {
it("null provider → the honest unavailability string, never a crash", async () => {
const { web_search } = createWebTools(() => null);
assert.ok(web_search, "web_search is still exposed without a provider");
const out = await web_search!.execute!("c1", { query: "anything" });
assert.match(String(out.details), /not available: no search engine is configured/);
assert.match(String(out.details), /Tavily or Brave/);
const missing = "web_search is not available: no search engine is configured. " +
"Set a Tavily or Brave search API key in settings. Report in your brief that web search is unavailable.";
assert.equal(out.details, missing);
assert.deepEqual(out.content, [{ type: "text", text: missing }]);
assert.notEqual(out.isError, true, "missing keys remain ordinary recoverable text, not a native error");
});
it("a live provider returns its hits (count defaults to 5)", async () => {
@@ -110,11 +108,86 @@ describe("web_search on the injected port (web.ts)", () => {
});
});
/** Minimal in-memory JobStore — the drop tests never touch the fs. */
describe("keyless researcher loop", () => {
it("sends the exact approved missing-credentials guidance in the researcher system prompt", async () => {
const approved = "If web_search reports that search is unavailable because credentials are missing, do not retry it with different queries. Use web_fetch when you have source URLs, and continue with any other available sources. If the missing search capability materially limits your work, state the limitation and the missing-key setup needed in your brief. Never invent search results or sources.";
const faux = createFauxCore({});
faux.setResponses([(context) => {
const system = context.messages.find((m) => m.role === "system");
assert.ok(system && system.role === "system");
assert.equal(typeof system.content, "string");
const prompt = system.content as string;
assert.ok(prompt.split("\n\n").includes(approved), "approved paragraph is passed verbatim");
assert.match(prompt, /Break the question into 2-4 searchable facets/);
assert.match(prompt, /## Gaps/);
assert.ok(system.toolsAdded?.some((t) => t.name === "web_search"));
assert.ok(system.toolsAdded?.some((t) => t.name === "web_fetch"));
return fauxAssistantMessage([fauxText("Prompt verified; no model behavior inferred.")]);
}]);
assert.equal(await runResearcherAgent("Check the prompt only", {
engine: { model: faux.getModel(), streamFn: faux.streamSimple },
}), "Prompt verified; no model behavior inferred.");
});
// These scripts prove tool/loop mechanics, not real-model compliance.
for (const searchFirst of [true, false]) {
it(searchFirst ? "recovers from missing search keys by fetching a supplied URL" : "fetch-only research needs no search credentials", async () => {
const url = "https://example.com/supplied-source";
const brief = `## Summary\nVerified supplied source. [Source](${url})`;
const steps: FauxResponseStep[] = [];
const fetchCall = () => fauxAssistantMessage(
[fauxToolCall("web_fetch", { url }, { id: "fetch1" })], { stopReason: "toolUse" },
);
if (searchFirst) {
steps.push(
() => {
return fauxAssistantMessage(
[fauxToolCall("web_search", { query: "supplied source" }, { id: "search1" })], { stopReason: "toolUse" },
);
},
(context) => {
const result = context.messages.find((m) => m.role === "toolResult" && m.toolCallId === "search1");
assert.ok(result && result.role === "toolResult");
assert.equal(result.isError, false, "missing keys do not abort the native loop");
assert.match(JSON.stringify(result.content), /no search engine is configured/);
return fetchCall();
},
);
} else steps.push(fetchCall());
steps.push((context) => {
const result = context.messages.find((m) => m.role === "toolResult" && m.toolCallId === "fetch1");
assert.ok(result && result.role === "toolResult");
assert.equal(result.isError, false);
assert.match(JSON.stringify(result.content), /Synthetic verified source content/);
return fauxAssistantMessage([fauxText(brief)]);
});
const faux = createFauxCore({});
faux.setResponses(steps);
const realFetch = globalThis.fetch;
const fetched: string[] = [];
globalThis.fetch = (async (input) => {
fetched.push(String(input));
assert.equal(String(input), url, "only the supplied URL is fetched");
return new Response("Synthetic verified source content", { headers: { "content-type": "text/plain" } });
}) as typeof fetch;
try {
assert.equal(await runResearcherAgent(`Verify ${url}`, {
engine: { model: faux.getModel(), streamFn: faux.streamSimple },
// No search seam: this is the keyless backstop, not a search stub.
}), brief);
assert.deepEqual(fetched, [url]);
} finally {
globalThis.fetch = realFetch;
}
});
}
});
/** Minimal in-memory JobStore — registry tests never touch the fs. */
function fakeJobStore(): JobStore {
const jobs = new Map<string, JobRecord>();
return {
createJob(init) {
createJob(init) {
const job: JobRecord = {
id: `job_${jobs.size + 1}`,
kind: init.kind,
@@ -149,13 +222,28 @@ function fakeCtx(available?: () => boolean) {
};
}
describe("the research registry drop (jobTools)", () => {
it("searchUnavailable: false → research drops from the registry (the other job tools stay)", () => {
describe("keyless research dispatch (jobTools)", () => {
it("searchAvailable: false → research remains in the registry", () => {
const tools = jobTools(fakeCtx(() => false));
assert.equal("research" in tools, false, "no research tool — no doomed dispatch");
assert.ok(tools.research, "research can fetch supplied URLs without a search provider");
assert.ok(tools.demo && tools.visual && tools.pdf, "the search-less server keeps its other sub-agents");
});
it("keyless dispatch preserves the async job acknowledgement and spawn contract", async () => {
const ctx = fakeCtx(() => false);
const spawned: Array<{ job: JobRecord; input: unknown }> = [];
const tools = jobTools({ ...ctx, spawn: (job, input) => { spawned.push({ job, input }); } });
const input = { question: "Verify the supplied URL", context: "https://example.com/source" };
const out = await tools.research!.execute!("keyless-call", input);
assert.match(String(out.details), /^Dispatched \(job job_1\)\./);
assert.match(String(out.details), /fire-and-forget/);
assert.equal(spawned.length, 1);
assert.equal(spawned[0]!.job.kind, "research");
assert.equal(spawned[0]!.job.toolCallId, "keyless-call");
assert.equal(spawned[0]!.job.status, "running");
assert.deepEqual(spawned[0]!.input, input);
});
it("searchAvailable: true → research present", () => {
const tools = jobTools(fakeCtx(() => true));
assert.ok(tools.research, "a keyed server exposes research");
@@ -167,12 +255,8 @@ describe("the research registry drop (jobTools)", () => {
});
});
// ─── the threaded gate (fix wave, 2026-09-20) ────────────────────────────────
//
// The fix: createTeacherTools/createProberTools used to DROP searchAvailable
// when building the inner JobToolContext, so jobTools' gate always took its
// ?? true default even when the live wiring said unavailable. These pins
// hold the THREADED path (the registry factories, not jobTools directly).
// Exercise live registry factories too: search status must not override
// the existing per-role/user-toggle policy.
import { createProberTools, createTeacherTools } from "../src/agent/tools.ts";
import type { ChatStore } from "../src/core/ports.ts";
@@ -186,19 +270,19 @@ const fakeChatStore = {
const ctxBase = { chatId: "chat-gate-test", chatStore: fakeChatStore };
describe("the threaded searchAvailable gate (registry factories)", () => {
it("createTeacherTools: unavailable drops research on the live wiring path", () => {
describe("keyless role registries", () => {
it("createTeacherTools: unavailable keeps research on the live wiring path", () => {
const tools = createTeacherTools({
...ctxBase,
jobStore: fakeJobStore(),
spawn: () => {},
searchAvailable: () => false,
});
assert.equal("research" in tools, false, "the threaded signal reached jobTools' gate");
assert.ok(tools.research, "no search key is required to dispatch research");
assert.ok(tools.demo && tools.visual && tools.pdf, "the other job tools survive");
});
it("createTeacherTools: available keeps research (the gate reads the signal both ways)", () => {
it("createTeacherTools: available keeps research", () => {
const tools = createTeacherTools({
...ctxBase,
jobStore: fakeJobStore(),
@@ -208,7 +292,7 @@ describe("the threaded searchAvailable gate (registry factories)", () => {
assert.ok(tools.research, "a keyed server exposes research through the teacher");
});
it("createProberTools: research + pdf are phase-omnipresent; the gate drops research only", () => {
it("createProberTools: research + pdf remain phase-omnipresent with or without search keys", () => {
const on = createProberTools({
...ctxBase,
deliver: () => {},
@@ -225,16 +309,25 @@ describe("the threaded searchAvailable gate (registry factories)", () => {
spawn: () => {},
searchAvailable: () => false,
});
assert.equal("research" in off, false, "no engine = no research dispatch in the prober either");
assert.ok(off.research, "keyless research dispatch is available in the prober too");
assert.ok(off.pdf && off.probe_done, "pdf is keyless — it survives; probe_done always-on");
});
it("createProberTools: the user toggle drops research independently of the gate", () => {
it("createProberTools: the user toggle still drops research without search keys", () => {
const tools = createProberTools(
{ ...ctxBase, deliver: () => {}, jobStore: fakeJobStore(), spawn: () => {}, searchAvailable: () => true },
{ ...ctxBase, deliver: () => {}, jobStore: fakeJobStore(), spawn: () => {}, searchAvailable: () => false },
{ research: false },
);
assert.equal("research" in tools, false, "toggle-filtered like the teacher side");
assert.ok(tools.pdf, "pdf unaffected by a research toggle");
});
it("createTeacherTools: the user toggle still drops research without search keys", () => {
const tools = createTeacherTools(
{ ...ctxBase, jobStore: fakeJobStore(), spawn: () => {}, searchAvailable: () => false },
{ research: false },
);
assert.equal("research" in tools, false);
assert.ok(tools.pdf && tools.demo && tools.visual);
});
});