pygienium/tests/all-integration.test.ts

/**
 * all-integration.test.ts — `/pygienium-all` end-to-end (task 14).
 *
 * Mirrors the spec scenario: run every registered check in sequence under one
 * resumable run-state and confirm the run completes with every check marked
 * `complete`. Uses the injectable fake agent runner (no model needed) and stub
 * checks whose `!write`/`!echo` task protocol produces deterministic artifacts.
 */
import { describe, expect, it, beforeEach, afterEach } from "bun:test";
import { mkdtemp, readFile, rm } from "node:fs/promises";
import { tmpdir } from "node:os";
import { join } from "node:path";
import {
	clearChecks,
	registerCheck,
	type CheckDefinition,
} from "../src/checks/registry.js";
import {
	setAgentRunner,
	resetAgentRunner,
	fakeAgentRunner,
	type AgentRunner,
} from "../src/agent-runner.js";
import { handleAllCommand, type PygieniumCtx } from "../src/commands.js";
import { loadRunState } from "../src/run-state.js";
import { canonicalChecksRoot } from "../src/export.js";

/** Stub check whose fake-runner task writes artifacts + echoes a line. */
function fakeCheck(name: string): CheckDefinition {
	return {
		name,
		label: name,
		description: `${name} check`,
		agentName: "scanner",
		fixAgentName: "fixer",
		phaseId: "scan",
		buildScanTask: (_cwd, scope) =>
			`!write pygienium/checks/${name}/findings.md # ${name} findings\nscan-target:${scope.target}\n!echo ${name}-scan`,
		buildFixTask: (_cwd, _scope, findings) =>
			`!write pygienium/checks/${name}/changes.md # ${name} changes\nbased-on:${findings.split("\n")[0] ?? ""}\n!echo ${name}-fix`,
		gate: () => undefined,
	};
}

function stubCtx(cwd: string): PygieniumCtx {
	return { cwd, mode: "print", hasUI: false, ui: undefined } as PygieniumCtx;
}

/** Capture process.stdout.write lines for the duration of `fn`. */
async function captureStdout(fn: () => Promise<void>): Promise<string[]> {
	const out: string[] = [];
	const write = process.stdout.write.bind(process.stdout);
	(process.stdout as { write: (chunk: unknown) => boolean }).write = (
		chunk: unknown,
	) => {
		out.push(String(chunk).replace(/\r?\n$/, ""));
		return true;
	};
	try {
		await fn();
	} finally {
		(process.stdout as { write: (chunk: unknown) => boolean }).write = write;
	}
	return out;
}

describe("/pygienium-all end-to-end (task 14)", () => {
	let cwd: string;
	let dispatched: string[];
	let runner: AgentRunner;

	beforeEach(async () => {
		clearChecks();
		dispatched = [];
		runner = async (opts) => {
			const m = /pygienium\/checks\/([^/]+)\//.exec(opts.task);
			if (m) dispatched.push(m[1] as string);
			return fakeAgentRunner(opts);
		};
		setAgentRunner(runner);
		cwd = await mkdtemp(join(tmpdir(), "pygienium-all-"));
	});

	afterEach(async () => {
		resetAgentRunner();
		await rm(cwd, { recursive: true, force: true });
	});

	it("runs every registered check in sequence and marks the run complete", async () => {
		registerCheck(fakeCheck("alpha"));
		registerCheck(fakeCheck("beta"));
		registerCheck(fakeCheck("gamma"));

		const out = await captureStdout(() =>
			handleAllCommand("--fix", stubCtx(cwd)),
		);

		// Every check was dispatched (scan + fix each, in registration order).
		expect(dispatched.filter((n) => n === "alpha").length).toBeGreaterThan(0);
		expect(dispatched.filter((n) => n === "beta").length).toBeGreaterThan(0);
		expect(dispatched.filter((n) => n === "gamma").length).toBeGreaterThan(0);

		// Final run-state: complete, every check complete, recon shared once.
		const state = await loadRunState(cwd);
		expect(state).toBeDefined();
		expect(state!.status).toBe("complete");
		expect(state!.recon.complete).toBe(true);
		for (const name of ["alpha", "beta", "gamma"]) {
			expect(state!.checks[name]?.status).toBe("complete");
		}

		// Per-check artifacts landed on disk under the canonical root.
		const alphaFindings = await readFile(
			join(canonicalChecksRoot(cwd), "alpha", "findings.md"),
			"utf8",
		);
		expect(alphaFindings).toContain("alpha findings");
		const gammaChanges = await readFile(
			join(canonicalChecksRoot(cwd), "gamma", "changes.md"),
			"utf8",
		);
		expect(gammaChanges).toContain("gamma changes");

		// The summary line reports completion and the run-state path.
		const text = out.join("\n");
		expect(text).toContain("pygienium: all-run complete");
	});

	it("completes cleanly in scan-only mode (no --fix)", async () => {
		registerCheck(fakeCheck("solo"));
		const out = await captureStdout(() => handleAllCommand("", stubCtx(cwd)));
		const state = await loadRunState(cwd);
		expect(state!.status).toBe("complete");
		expect(state!.checks["solo"]?.status).toBe("complete");
		expect(state!.checks["solo"]?.fix).toBe(false);
		// Scan-only still writes findings but not changes.
		const findings = await readFile(
			join(canonicalChecksRoot(cwd), "solo", "findings.md"),
			"utf8",
		);
		expect(findings).toContain("solo findings");
		expect(out.join("\n")).toContain("pygienium: all-run complete");
	});

	it("reports no checks when the registry is empty", async () => {
		const out = await captureStdout(() => handleAllCommand("", stubCtx(cwd)));
		expect(out.join("\n")).toContain("no checks registered");
	});

	it("is resumable: a second call reuses the existing run-state", async () => {
		registerCheck(fakeCheck("alpha"));
		await captureStdout(() => handleAllCommand("--fix", stubCtx(cwd)));
		const first = await loadRunState(cwd);
		const firstStarted = first!.startedAt;

		// Second run reloads the existing run-state (same startedAt).
		await captureStdout(() => handleAllCommand("--fix", stubCtx(cwd)));
		const second = await loadRunState(cwd);
		expect(second!.startedAt).toBe(firstStarted);
		expect(second!.status).toBe("complete");
	});
});