plugin-legal-ai/scripts/int9-guard.mjs

#!/usr/bin/env node
/**
 * INV-INT9 guard — enforce single write-ownership of `issue.status`
 * (legal-ai docs/spec/X7-paperclip-client-params.md §4 INV-INT9).
 *
 * `issues.status` has exactly one legitimate writer in this plugin: the
 * `sync-case-status` scheduled job (`ctx.jobs.register("sync-case-status", …)`
 * in `src/worker.ts`), which mirrors legal-ai's case status onto the linked
 * Paperclip issue. Any OTHER `ctx.issues.update(...)` call that touches
 * `status` is a second writer racing the sync job — exactly the flip-flop
 * (`done` → `in_progress`) bug class legal-ai issue #618 exists to close off
 * for good, before a second call site is ever added.
 *
 * IN SCOPE: every `.ts`/`.tsx` file under `src/` (including `*.test.ts` —
 * there is no false positive there today, and the noqa escape hatch below
 * covers the day there is one).
 *
 * OUT OF SCOPE: this is NOT a general call-graph analysis. It is a
 * structural, single-pass lexer over the source text — no type information,
 * no cross-file resolution. It answers exactly one question per call site:
 * "does the character range of this `ctx.issues.update(...)` call sit
 * inside the character range of the one whitelisted
 * `ctx.jobs.register("sync-case-status", …)` call?". Deliberately NOT a
 * per-file allowlist (the pattern `scripts/leak_guard.py` in legal-ai uses)
 * — that shape does not fit here, because the one legitimate call site and
 * every future illegitimate one live in the very same file (`worker.ts`).
 * The granularity has to be a code RANGE derived from the real structure,
 * not a file name.
 *
 * Escape hatch (G1 — fix at the source, don't paper over a false positive):
 * a call can be justified in place with
 *     // noqa: INT9 — <reason, at least 3 characters>
 * on the line immediately above the call, or on any line spanned by its
 * argument list. A bare `// noqa: INT9` with no reason is itself a
 * violation (AC3 — exceptions must be justified in writing, not silent).
 *
 * Usage:
 *   node scripts/int9-guard.mjs                 # scan src/**\/*.{ts,tsx}; exit 1 on any violation
 *   node scripts/int9-guard.mjs <path>...        # scan only the given files/directories
 *   node scripts/int9-guard.mjs --self-test      # run the fixture suite (scripts/fixtures/int9/), prove the gate bites
 *
 * Zero dependencies (node:fs / node:path / node:process only) — no `npm ci`
 * needed in CI; the runner image ships node 24.
 */
import { readdirSync, readFileSync, statSync } from "node:fs";
import { join, relative, resolve } from "node:path";
import process from "node:process";

const REPO_ROOT = resolve(import.meta.dirname, "..");
const SKIP_DIRS = new Set(["node_modules", "dist", ".git"]);
const SOURCE_EXT = new Set([".ts", ".tsx"]);

const SPEC_REF = "docs/spec/X7-paperclip-client-params.md §4 INV-INT9";

const NOQA_RE = /\/\/\s*noqa\s*:\s*INT9\b(?:\s*[-—]\s*(.*))?/i;

const REGEX_PREV_CHARS = new Set("(,=:[!&|?{};+-*%~^<>".split(""));
const REGEX_PREV_WORDS = new Set([
	"return",
	"typeof",
	"case",
	"in",
	"of",
	"new",
	"delete",
	"void",
	"instanceof",
	"do",
	"else",
	"yield",
	"await",
]);

// ---------------------------------------------------------------------
// Tokenizer — classify every character as "plain code" or not, so that
// structural regexes and paren-matching never see string/template
// contents, comments, or regex-literal bodies.
// ---------------------------------------------------------------------

/** @returns {Uint8Array} mask[i] === 1 iff text[i] is plain code. */
function classify(text) {
	const n = text.length;
	const mask = new Uint8Array(n);
	/** @type {Array<{kind: "template"} | {kind: "exprCode", depth: number}>} */
	const stack = [];
	let state = "code";
	let lastSignificant = "";
	let lastWord = "";
	let i = 0;

	function noteCodeChar(ch) {
		if (/\s/.test(ch)) return;
		lastSignificant = ch;
		if (/[A-Za-z0-9_$]/.test(ch)) {
			lastWord += ch;
		} else {
			lastWord = "";
		}
	}

	while (i < n) {
		const ch = text[i];

		if (state === "code") {
			const next = text[i + 1];
			if (ch === "/" && next === "/") {
				state = "lcomment";
				i += 2;
				continue;
			}
			if (ch === "/" && next === "*") {
				state = "bcomment";
				i += 2;
				continue;
			}
			if (ch === "'") {
				state = "squote";
				i += 1;
				continue;
			}
			if (ch === '"') {
				state = "dquote";
				i += 1;
				continue;
			}
			if (ch === "`") {
				stack.push({ kind: "template" });
				state = "template";
				i += 1;
				continue;
			}
			if (ch === "/") {
				const isRegex =
					lastSignificant === "" ||
					REGEX_PREV_CHARS.has(lastSignificant) ||
					REGEX_PREV_WORDS.has(lastWord);
				if (isRegex) {
					state = "regex";
					i += 1;
					continue;
				}
				// else: division — falls through to plain-code handling below.
			}

			mask[i] = 1;
			const top = stack[stack.length - 1];
			if (ch === "{" && top && top.kind === "exprCode") {
				top.depth += 1;
			} else if (ch === "}" && top && top.kind === "exprCode") {
				top.depth -= 1;
				if (top.depth === 0) {
					stack.pop();
					state = "template";
				}
			}
			noteCodeChar(ch);
			i += 1;
			continue;
		}

		if (state === "squote" || state === "dquote") {
			const quote = state === "squote" ? "'" : '"';
			if (ch === "\\") {
				i += 2;
				continue;
			}
			if (ch === quote) {
				state = "code";
				noteCodeChar(quote);
				i += 1;
				continue;
			}
			if (ch === "\n") {
				// Unterminated string — bail back to code rather than eat the file.
				state = "code";
				i += 1;
				continue;
			}
			i += 1;
			continue;
		}

		if (state === "template") {
			if (ch === "\\") {
				i += 2;
				continue;
			}
			if (ch === "`") {
				stack.pop();
				state = "code";
				i += 1;
				continue;
			}
			if (ch === "$" && text[i + 1] === "{") {
				stack.push({ kind: "exprCode", depth: 1 });
				state = "code";
				i += 2;
				continue;
			}
			i += 1;
			continue;
		}

		if (state === "lcomment") {
			if (ch === "\n") {
				state = "code";
				i += 1;
				continue;
			}
			i += 1;
			continue;
		}

		if (state === "bcomment") {
			if (ch === "*" && text[i + 1] === "/") {
				state = "code";
				i += 2;
				continue;
			}
			i += 1;
			continue;
		}

		if (state === "regex") {
			if (ch === "\\") {
				i += 2;
				continue;
			}
			if (ch === "[") {
				state = "regexclass";
				i += 1;
				continue;
			}
			if (ch === "/") {
				i += 1;
				while (i < n && /[a-z]/i.test(text[i])) i += 1;
				state = "code";
				continue;
			}
			if (ch === "\n") {
				// Unterminated regex — bail back to code rather than eat the file.
				state = "code";
				i += 1;
				continue;
			}
			i += 1;
			continue;
		}

		if (state === "regexclass") {
			if (ch === "\\") {
				i += 2;
				continue;
			}
			if (ch === "]") {
				state = "regex";
				i += 1;
				continue;
			}
			i += 1;
			continue;
		}

		// Unreachable, but never silently drop a character.
		i += 1;
	}

	return mask;
}

/** Yield regex matches whose full span lies entirely in plain-code region. */
function* matchAllCode(text, mask, pattern) {
	for (const m of text.matchAll(pattern)) {
		const start = m.index;
		const end = start + m[0].length;
		let allCode = true;
		for (let i = start; i < end; i++) {
			if (!mask[i]) {
				allCode = false;
				break;
			}
		}
		if (allCode) yield m;
	}
}

/** Index of the `)` matching the `(` at `openIndex`, counting only plain-code parens. */
function matchParen(text, mask, openIndex, filePath) {
	let depth = 0;
	for (let i = openIndex; i < text.length; i++) {
		if (!mask[i]) continue;
		if (text[i] === "(") depth += 1;
		else if (text[i] === ")") {
			depth -= 1;
			if (depth === 0) return i;
		}
	}
	throw new Error(`${filePath}: פענוח נכשל — סוגר לא נסגר`);
}

function buildLineStarts(text) {
	const starts = [0];
	for (let i = 0; i < text.length; i++) {
		if (text[i] === "\n") starts.push(i + 1);
	}
	return starts;
}

function lineOfOffset(lineStarts, idx) {
	let lo = 0;
	let hi = lineStarts.length - 1;
	while (lo < hi) {
		const mid = (lo + hi + 1) >> 1;
		if (lineStarts[mid] <= idx) lo = mid;
		else hi = mid - 1;
	}
	return lo; // 0-based line number
}

const STATUS_WORD_RE = /\bstatus\b/;
const JOBS_REGISTER_RE = /jobs\.register\s*\(/g;
const ISSUES_UPDATE_RE = /issues\.update\s*\(/g;
const SYNC_CASE_STATUS_LITERAL_RE = /^\s*["']sync-case-status["']/;

/**
 * Scan one file's source text and return a list of violation objects
 * ({ file, line, message }). Throws if a `(` is never closed (see
 * matchParen) — the caller must NOT treat that as "no violations".
 */
function scanFile(filePath, text) {
	const mask = classify(text);
	const lines = text.split("\n");
	const lineStarts = buildLineStarts(text);
	const rel = filePath;

	const allowedRanges = [];
	for (const m of matchAllCode(text, mask, JOBS_REGISTER_RE)) {
		const openIdx = m.index + m[0].length - 1;
		const after = text.slice(openIdx + 1, openIdx + 60);
		if (SYNC_CASE_STATUS_LITERAL_RE.test(after)) {
			const closeIdx = matchParen(text, mask, openIdx, rel);
			allowedRanges.push([openIdx, closeIdx]);
		}
	}

	const violations = [];
	for (const m of matchAllCode(text, mask, ISSUES_UPDATE_RE)) {
		const openIdx = m.index + m[0].length - 1;
		const closeIdx = matchParen(text, mask, openIdx, rel);
		const content = text.slice(openIdx + 1, closeIdx);
		if (!STATUS_WORD_RE.test(content)) continue; // not a status-write call

		const inAllowedRange = allowedRanges.some(
			([s, e]) => openIdx >= s && openIdx <= e,
		);
		if (inAllowedRange) continue;

		const callStartLine = lineOfOffset(lineStarts, m.index);
		const callEndLine = lineOfOffset(lineStarts, closeIdx);

		let noqaMatch;
		for (let ln = callStartLine - 1; ln <= callEndLine && !noqaMatch; ln += 1) {
			if (ln < 0 || ln >= lines.length) continue;
			const m2 = NOQA_RE.exec(lines[ln]);
			if (m2) noqaMatch = m2;
		}

		const reportLine = callStartLine + 1; // 1-based for humans
		if (noqaMatch) {
			const reason = (noqaMatch[1] ?? "").trim();
			if (reason.length >= 3) continue; // justified exception — allowed
			violations.push({
				file: rel,
				line: reportLine,
				message: "noqa: INT9 ללא נימוק — חריגה חייבת להיות מנומקת בכתב (AC3)",
			});
			continue;
		}

		violations.push({
			file: rel,
			line: reportLine,
			message:
				`issues.update({status}) מחוץ למסלול sync-case-status (INV-INT9). ` +
				`ראה ${SPEC_REF}. חריגה מוצדקת → // noqa: INT9 — <נימוק>.`,
		});
	}

	return violations;
}

// ---------------------------------------------------------------------
// File collection
// ---------------------------------------------------------------------

function walk(dir, out) {
	for (const entry of readdirSync(dir, { withFileTypes: true })) {
		if (SKIP_DIRS.has(entry.name)) continue;
		const full = join(dir, entry.name);
		if (entry.isDirectory()) {
			walk(full, out);
		} else if (SOURCE_EXT.has(entry.name.slice(entry.name.lastIndexOf(".")))) {
			out.push(full);
		}
	}
	return out;
}

function collectDefault() {
	const srcDir = join(REPO_ROOT, "src");
	return walk(srcDir, []);
}

function collectFromArgs(args) {
	const out = [];
	for (const a of args) {
		const p = resolve(a);
		const st = statSync(p);
		if (st.isDirectory()) {
			walk(p, out);
		} else {
			out.push(p);
		}
	}
	return out;
}

function runGate(files) {
	const violations = [];
	for (const abs of files) {
		const rel = relative(REPO_ROOT, abs);
		const text = readFileSync(abs, "utf-8");
		violations.push(...scanFile(rel, text));
	}
	return violations;
}

function reportGate(violations, scannedCount) {
	if (violations.length > 0) {
		process.stderr.write(
			`✗ INV-INT9 gate — issue.status write-ownership violated ` +
				`(${violations.length} finding(s)):\n\n`,
		);
		for (const v of violations) {
			process.stderr.write(`  • ${v.file}:${v.line}: ${v.message}\n`);
		}
		process.stderr.write(`\nSee ${SPEC_REF}.\n`);
		return 1;
	}
	process.stdout.write(
		`✓ INV-INT9 gate: מסלול-כתיבה יחיד ל-issue.status ` +
			`(${scannedCount} קבצים נסרקו).\n`,
	);
	return 0;
}

// ---------------------------------------------------------------------
// Self-test
// ---------------------------------------------------------------------

const SELF_TEST_EXPECTED = {
	"violating.ts": 1,
	"allowed-inside-job.ts": 0,
	"noqa-justified.ts": 0,
	"noqa-bare.ts": 1,
	"non-status-update.ts": 0,
};

function runSelfTest() {
	const fixturesDir = join(REPO_ROOT, "scripts", "fixtures", "int9");
	let pass = 0;
	let total = 0;
	const failures = [];

	for (const [name, expected] of Object.entries(SELF_TEST_EXPECTED)) {
		total += 1;
		const abs = join(fixturesDir, name);
		let violations;
		try {
			const text = readFileSync(abs, "utf-8");
			violations = scanFile(name, text);
		} catch (err) {
			failures.push(
				`${name}: ציפינו ${expected} הפרות, אבל הסריקה זרקה שגיאה: ${String(err)}`,
			);
			continue;
		}
		if (violations.length === expected) {
			pass += 1;
		} else {
			failures.push(
				`${name}: ציפינו ${expected} הפרות, התקבלו ${violations.length}` +
					(violations.length
						? `:\n${violations.map((v) => `      - ${v.file}:${v.line}: ${v.message}`).join("\n")}`
						: ""),
			);
		}
	}

	if (failures.length > 0) {
		process.stderr.write("✗ INV-INT9 self-test נכשל:\n\n");
		for (const f of failures) process.stderr.write(`  • ${f}\n`);
		return 1;
	}
	process.stdout.write(
		`✓ INV-INT9 self-test: ${pass}/${total} פיקסצ'רים תואמים לציפייה.\n`,
	);
	return 0;
}

// ---------------------------------------------------------------------
// Entry point
// ---------------------------------------------------------------------

function main(argv) {
	if (argv.includes("--self-test")) {
		return runSelfTest();
	}
	const pathArgs = argv.filter((a) => !a.startsWith("--"));
	const files =
		pathArgs.length > 0 ? collectFromArgs(pathArgs) : collectDefault();
	let violations;
	try {
		violations = runGate(files);
	} catch (err) {
		process.stderr.write(`✗ INV-INT9 gate — ${String(err.message ?? err)}\n`);
		return 1;
	}
	return reportGate(violations, files.length);
}

// `exitCode` ולא `process.exit()` — ‏stdout/stderr אל pipe (וזה בדיוק המצב ב-CI)
// אסינכרוניים ב-POSIX, ו-`process.exit` היה עלול לקטוע את דוח-ההפרות עצמו.
process.exitCode = main(process.argv.slice(2));