// Corpus guard: authored OpenProse intent stays independent of harness products.
//
// The contracts in this repository are harness-agnostic. The reference
// harness (Reactor: `@openprose/reactor`, `@openprose/reactor-cli`,
// `@openprose/reactor-devtools`, the `reactor` binary) lives in its own
// repository, and the run-phase model is named for what it is — the
// reconciler — not for one implementation of it. This test keeps it that way,
// the same way stale-docs.test.ts keeps retired kinds from creeping back in.
//
// Two kinds of token are checked across the skill, the spec, the std/co contract
// libraries, and the root docs:
//   - PRODUCT tokens (package names, the CLI, its config and env vars) — a
//     line carrying one FAILS unless it sits in an allowlisted pointer.
//   - The MODEL token (the bare word "Reactor" / "reactor"). The run-phase
//     model is "the reconciler"; the spec's harness contract is "a conforming
//     harness".
//
// Allowlisted pointers — the two places the product is intentionally named:
//   - README.md, the lines under its `## Harnesses` heading.
//   - skills/open-prose/changelog.md, the upgrade record (it must name what
//     moved so `prose upgrade` can route users).
//
// The example corpus is a one-way boundary: it owns harness-neutral authored
// intent and conformance expectations, while implementations prove conformance
// independently. Reactor implementation links, product/model prose, and product
// walkthroughs are forbidden across every file under examples/.
//
// Separately, the 16 legacy expanded-topology READMEs named below must not teach
// unsupported native compile/promote/serve/status recipes. That narrower check is
// intentionally not applied to native-ready examples.
//
// Doc-conformance style: read the docs off disk, assert on content, no runtime.
// The repo-root vitest config discovers tests/open-prose/**/*.test.ts, so
// `pnpm test` picks this up.
import { existsSync, readdirSync, readFileSync, statSync } from "node:fs";
import { join, relative } from "node:path";
import { fileURLToPath } from "node:url";
import { describe, expect, it } from "vitest";

const repoRoot = fileURLToPath(new URL("../../../", import.meta.url));
const examplesRoot = join(repoRoot, "skills/open-prose/examples");

// Directories walked for every *.md (which includes *.prose.md). The skill
// root covers its example corpus too.
const ROOTS = ["skills/open-prose", "spec", "packages/std", "packages/co"];
// Root-level docs checked as single files.
const ROOT_FILES = ["README.md", "CONTRIBUTING.md", "AGENTS.md"];

// Legacy expanded-topology examples whose authored intent predates native run
// support. Keep this scope explicit; other examples may document native recipes.
const LEGACY_EXAMPLE_READMES = [
	"skills/open-prose/examples/agent-observatory/README.md",
	"skills/open-prose/examples/basic-unit-suite/README.md",
	"skills/open-prose/examples/feedback-pulse/README.md",
	"skills/open-prose/examples/forme-fixpoint/README.md",
	"skills/open-prose/examples/github-star-enricher/README.md",
	"skills/open-prose/examples/implementation-pipeline/README.md",
	"skills/open-prose/examples/inbox-triage/README.md",
	"skills/open-prose/examples/masked-relay/README.md",
	"skills/open-prose/examples/monorepo-ci/README.md",
	"skills/open-prose/examples/oblique-weave/README.md",
	"skills/open-prose/examples/press-desk/README.md",
	"skills/open-prose/examples/renewal-risk/README.md",
	"skills/open-prose/examples/research-tree/README.md",
	"skills/open-prose/examples/support-inbox-router/README.md",
	"skills/open-prose/examples/surprise-cost/README.md",
	"skills/open-prose/examples/tamper-forge/README.md",
];

const UNSUPPORTED_NATIVE_RUN_RECIPE_TOKENS = [
	"prose compile",
	"manifest.next.json",
	"prose promote",
	"prose serve",
	"prose status",
];

// The product: packages, binary, config, env vars, the retired skill verb and
// operator guide, and source paths that only exist in the harness repository.
const PRODUCT_TOKENS: RegExp[] = [
	/@openprose\/reactor/,
	/\breactor-cli\b/,
	/\breactor-devtools\b/,
	/\breactor\.yml\b/,
	/\breactor\.md\b/,
	/\bprose react\b/,
	/packages\/reactor/,
	/\bREACTOR_[A-Z_]+\b/,
	/\bnpx reactor\b/,
];

// The model: the bare product name used for the run-phase reconciler or the
// harness class. Case-insensitive so "the reactor DAG" is caught too.
const MODEL_TOKEN = /\breactor\b/i;
const PRODUCT_GITHUB_URL = /github\.com\/openprose\/reactor/i;

// Files whose job is to record the migration.
const ALLOWLIST_FILES = new Set(["skills/open-prose/changelog.md"]);
// Files allowed one section: lines from this H2 up to the next H2.
const ALLOWLIST_SECTIONS: Record<string, string> = { "README.md": "## Harnesses" };

function allDocs(dir: string): string[] {
	const out: string[] = [];
	for (const entry of readdirSync(dir)) {
		const full = join(dir, entry);
		if (statSync(full).isDirectory()) {
			out.push(...allDocs(full));
		} else if (entry.endsWith(".md")) {
			out.push(full);
		}
	}
	return out;
}

function allFiles(dir: string): string[] {
	const out: string[] = [];
	for (const entry of readdirSync(dir)) {
		const full = join(dir, entry);
		if (statSync(full).isDirectory()) out.push(...allFiles(full));
		else out.push(full);
	}
	return out;
}

function scanTargets(): string[] {
	const files: string[] = [];
	for (const root of ROOTS) files.push(...allDocs(join(repoRoot, root)));
	for (const file of ROOT_FILES) {
		const full = join(repoRoot, file);
		if (existsSync(full)) files.push(full);
	}
	return files;
}

// Line numbers (1-based) that sit under the allowlisted H2 of a file, if any.
function allowlistedLines(rel: string, lines: string[]): Set<number> {
	const heading = ALLOWLIST_SECTIONS[rel];
	const allowed = new Set<number>();
	if (!heading) return allowed;
	let inside = false;
	lines.forEach((line, i) => {
		if (line.startsWith("## ")) inside = line.trim() === heading;
		if (inside) allowed.add(i + 1);
	});
	return allowed;
}

type Offender = string;

function collect(matches: (line: string) => boolean): Offender[] {
	const offenders: Offender[] = [];
	for (const path of scanTargets()) {
		const rel = relative(repoRoot, path);
		if (ALLOWLIST_FILES.has(rel)) continue;
		const lines = readFileSync(path, "utf8").split("\n");
		const allowed = allowlistedLines(rel, lines);
		lines.forEach((line, i) => {
			if (!matches(line)) return;
			if (allowed.has(i + 1)) return;
			offenders.push(`${rel}:${i + 1} — ${line.trim()}`);
		});
	}
	return offenders;
}

describe("harness surface — the OpenProse repo does not teach the Reactor product", () => {
	it("finds docs to scan, the example corpus included (sanity)", () => {
		const targets = scanTargets().map((path) => relative(repoRoot, path));
		expect(targets.length).toBeGreaterThan(20);
		expect(targets).toContain("skills/open-prose/examples/README.md");
		expect(targets.filter((rel) => rel.startsWith("skills/open-prose/examples/")).length)
			.toBeGreaterThan(30);
	});

	it("no OpenProse surface teaches the Reactor product as live", () => {
		const offenders = collect((line) => PRODUCT_TOKENS.some((token) => token.test(line)));
		expect(offenders, `product surface taught as live:\n${offenders.join("\n")}`).toEqual([]);
	});

	it("named legacy expanded-topology READMEs omit unsupported native recipes", () => {
		const offenders = LEGACY_EXAMPLE_READMES.flatMap((readme) => {
			const content = readFileSync(join(repoRoot, readme), "utf8");
			return UNSUPPORTED_NATIVE_RUN_RECIPE_TOKENS.filter((token) => content.includes(token)).map(
				(token) => `${readme} — ${token}`,
			);
		});

		expect(
			offenders,
			`unsupported native run-recipe tokens:\n${offenders.join("\n")}`,
		).toEqual([]);
	});

	it("example files contain no product URL or bare product token", () => {
		const offenders = allFiles(examplesRoot).flatMap((path) => {
			const content = readFileSync(path, "utf8");
			return content
				.split("\n")
				.map((line, i) => ({ line, number: i + 1 }))
				.filter(({ line }) => PRODUCT_GITHUB_URL.test(line) || MODEL_TOKEN.test(line))
				.map(({ line, number }) =>
					`${relative(repoRoot, path)}:${number} — ${line.trim()}`,
				);
		});

		expect(offenders, `product references in examples:\n${offenders.join("\n")}`).toEqual([]);
	});

	it("the run-phase model is not named after the product", () => {
		const offenders = collect((line) => MODEL_TOKEN.test(line));
		expect(offenders, `run-phase model named after the product:\n${offenders.join("\n")}`).toEqual([]);
	});

	it("the skill has no product-named operator guide or concept doc", () => {
		// cf. tenets.test.ts on runtime/judge-responsibility.prose.md: the retired
		// operator guide (`reactor.md`) and the product-named concept doc are gone;
		// the run-phase model lives in concepts/reconciler.md.
		const skillDir = join(repoRoot, "skills/open-prose");
		const productNamed = (dir: string) =>
			readdirSync(dir).filter((entry) => MODEL_TOKEN.test(entry));
		expect(productNamed(skillDir)).toEqual([]);
		expect(productNamed(join(skillDir, "concepts"))).toEqual([]);
		expect(existsSync(join(skillDir, "concepts/reconciler.md"))).toBe(true);
	});

	it("the spec has no product-named document", () => {
		const spec = readdirSync(join(repoRoot, "spec"));
		expect(spec.filter((entry) => MODEL_TOKEN.test(entry))).toEqual([]);
		expect(spec).toContain("02-Harness.md");
		expect(spec).toContain("03-AuthoringPattern.md");
	});

	it("the README keeps its Harnesses pointer", () => {
		// The one intentional home for the product name outside the changelog.
		const readme = readFileSync(join(repoRoot, "README.md"), "utf8");
		expect(readme).toContain("## Harnesses");
		expect(readme).toMatch(/github\.com\/openprose\/reactor/);
	});
});