omp-pygienium/tests/per-check-resume.test.ts

/**
 * per-check-resume.test.ts — `/pygienium-<check>` resume semantics.
 *
 * The per-check command must be resume-aware (parity with
 * `/pygienium-all` and `/pygienium-resume`): a check already terminal
 * (`complete`/`skipped`) is skipped unless `--fresh`, and a failed check is
 * re-dispatched. This is what lets "running the original command again"
 * recover a partial run instead of blindly re-running every phase.
 *
 * A stateful tracker wraps the fake runner: the first call produces no
 * artifact (verify fails), the second writes findings.md (verify passes) —
 * modelling the intermittent empty-output bug recovering on retry.
 */
import { describe, expect, it, beforeEach, afterEach } from "bun:test";
import { mkdtemp, mkdir, rm, writeFile, readFile } 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 { handleCheckCommand, type PygieniumCtx } from "../src/commands.js";
import { loadRunState } from "../src/run-state.js";

function stubCtx(cwd: string): PygieniumCtx {
	return { cwd,  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;
}

/** Check whose fake runner writes findings.md and (with --fix) changes.md. */
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 ${name}-fix`,
		// Verify hook asserting findings.md exists — like the real checks.
		verify: async (scope) => {
			const { stat } = await import("node:fs/promises");
			const f = join(scope.cwd, ".pygienium", "checks", name, "findings.md");
			try {
				await stat(f);
			} catch {
				return `${name} verify: expected findings.md at ${f} after scan, none found.`;
			}
			return undefined;
		},
		gate: () => undefined,
	};
}

/** Tracker: records dispatched agent tasks then delegates to the fake runner. */
function trackingRunner(): { runner: AgentRunner; dispatched: string[] } {
	const dispatched: string[] = [];
	const runner: AgentRunner = async (opts) => {
		const m = /pygienium\/checks\/([^/]+)\//.exec(opts.task);
		if (m) dispatched.push(m[1] as string);
		return fakeAgentRunner(opts);
	};
	return { runner, dispatched };
}

/**
 * Stateful runner: the first call returns ok with no artifact (verify fails),
 * the second writes findings.md (verify passes). Models the empty-output bug
 * recovering on retry.
 */
function flakyThenOkRunner(name: string): {
	runner: AgentRunner;
	calls: number;
} {
	let calls = 0;
	const runner: AgentRunner = async (opts) => {
		calls++;
		if (calls === 1) {
			return { ok: true, text: "" };
		}
		return fakeAgentRunner(opts);
	};
	return { runner, calls: 0 };
}

describe("/pygienium-<check> resume semantics", () => {
	let cwd: string;

	beforeEach(async () => {
		clearChecks();
		cwd = await mkdtemp(join(tmpdir(), "pygienium-pcr-"));
	});

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

	it("skips an already-complete check and does not re-dispatch the agent", async () => {
		const track = trackingRunner();
		setAgentRunner(track.runner);
		const check = fakeCheck("alpha");
		registerCheck(check);

		// First run: completes and writes findings.md.
		await handleCheckCommand(check, "", stubCtx(cwd));
		expect(track.dispatched).toEqual(["alpha"]);
		const state1 = await loadRunState(cwd);
		expect(state1?.checks.alpha.status).toBe("complete");

		// Second run: terminal → skipped, no agent dispatch.
		const out = await captureStdout(() =>
			handleCheckCommand(check, "", stubCtx(cwd)),
		);
		expect(track.dispatched).toEqual(["alpha"]); // unchanged
		expect(out.join("\n")).toContain("already complete");
		expect(out.join("\n")).toContain("--fresh");
		const state2 = await loadRunState(cwd);
		expect(state2?.checks.alpha.status).toBe("complete");
	});

	it("--fresh re-runs a completed check from scratch", async () => {
		const track = trackingRunner();
		setAgentRunner(track.runner);
		const check = fakeCheck("beta");
		registerCheck(check);

		await handleCheckCommand(check, "", stubCtx(cwd));
		expect(track.dispatched).toEqual(["beta"]);

		await captureStdout(() =>
			handleCheckCommand(check, "--fresh", stubCtx(cwd)),
		);
		// Dispatched again (now twice total).
		expect(track.dispatched).toEqual(["beta", "beta"]);
		const state = await loadRunState(cwd);
		expect(state?.checks.beta.status).toBe("complete");
	});

	it("re-runs a failed check and recovers when the agent produces the artifact on retry", async () => {
		const flaky = flakyThenOkRunner("gamma");
		// Expose the live call count via closure read after the run.
		let calls = 0;
		const runner: AgentRunner = async (opts) => {
			calls++;
			if (calls === 1) {
				return { ok: true, text: "" };
			}
			return fakeAgentRunner(opts);
		};
		void flaky; // (flakyThenOkRunner kept as a reference shape; use `runner` below)
		setAgentRunner(runner);

		const check = fakeCheck("gamma");
		registerCheck(check);

		// First run: agent returns ok with no artifact → verify fails.
		await handleCheckCommand(check, "", stubCtx(cwd));
		const state1 = await loadRunState(cwd);
		expect(state1?.checks.gamma.status).toBe("failed");
		expect(state1?.checks.gamma.error).toContain("verify");
		expect(state1?.checks.gamma.error).toContain("findings.md");

		// Resume: re-dispatch the failed check. Agent writes findings.md this
		// time → verify passes → complete.
		await captureStdout(() => handleCheckCommand(check, "", stubCtx(cwd)));
		const state2 = await loadRunState(cwd);
		expect(state2?.checks.gamma.status).toBe("complete");
		expect(state2?.checks.gamma.error).toBeUndefined();
		// The verify phase is now complete (not failed).
		const verify = state2?.checks.gamma.phases.find((p) => p.id === "verify");
		expect(verify?.status).toBe("complete");
		// And the artifact exists on disk.
		const findings = await readFile(
			join(cwd, ".pygienium", "checks", "gamma", "findings.md"),
			"utf8",
		);
		expect(findings).toContain("gamma findings");
	});

	it("does not treat a failed check as terminal (resume re-dispatches it)", async () => {
		const track = trackingRunner();
		setAgentRunner(track.runner);
		const check = fakeCheck("delta");
		// Override verify to always fail so the check lands in `failed`.
		const alwaysFailing: CheckDefinition = {
			...check,
			name: "delta",
			verify: async () => "delta verify: forced failure",
		};
		registerCheck(alwaysFailing);

		await handleCheckCommand(alwaysFailing, "", stubCtx(cwd));
		expect(track.dispatched).toEqual(["delta"]);
		const state1 = await loadRunState(cwd);
		expect(state1?.checks.delta.status).toBe("failed");

		// Re-running the command re-dispatches (failed is NOT terminal).
		await captureStdout(() =>
			handleCheckCommand(alwaysFailing, "", stubCtx(cwd)),
		);
		expect(track.dispatched).toEqual(["delta", "delta"]);
		const state2 = await loadRunState(cwd);
		expect(state2?.checks.delta.status).toBe("failed"); // still failing
	});
});

void mkdir; // silence unused-import lint under some configs