@oh-my-pi-zen/pi-coding-agent 16.3.6-zen.1

package/src/tools/path-utils.ts

@@ -0,0 +1,1279 @@
+import * as fs from "node:fs";
+import * as os from "node:os";
+import * as path from "node:path";
+import * as url from "node:url";
+import { isEnoent, isEnotdir, stripWindowsExtendedLengthPathPrefix } from "@oh-my-pi-zen/pi-utils";
+import type { Skill } from "../extensibility/skills";
+import { InternalUrlRouter, type LocalProtocolOptions } from "../internal-urls";
+import { ToolError } from "./tool-errors";
+
+const UNICODE_SPACES = /[\u00A0\u2000-\u200A\u202F\u205F\u3000]/g;
+// A single line-range chunk: `N`, `N-M`, `N+K`, or open-ended `N-`. `..` is
+// accepted everywhere `-` is, as a forgiving alias for Rust/Python-style ranges
+// (e.g. `2724..2727` == `2724-2727`, `2724..` == `2724-`); it is normalized to
+// `-` in parseLineRangeChunk. Keep this fragment and LINE_RANGE_CHUNK_RE in sync.
+const RANGE_CHUNK_SRC = String.raw`L?\d+(?:(?:[-+]|\.\.)L?\d+|-|\.\.)?`;
+const RANGE_LIST_SRC = `${RANGE_CHUNK_SRC}(?:,${RANGE_CHUNK_SRC})*`;
+const FILE_LINE_RANGE_RE = new RegExp(`^(?:${RANGE_LIST_SRC}|raw|conflicts)$`, "i");
+const FILE_LINE_RANGE_ONLY_RE = new RegExp(`^${RANGE_LIST_SRC}$`, "i");
+const FILE_RAW_ONLY_RE = /^raw$/i;
+// Permissive selector chunk for internal URLs — accepts well-formed selectors
+// plus common malformed shapes (e.g. `:-N`) so the read tool peels the entire
+// selector chain off before dispatching to a protocol handler.
+const INTERNAL_URL_SELECTOR_PART_RE = new RegExp(
+	String.raw`^(?:raw|conflicts|${RANGE_LIST_SRC}|-\d+(?:[-+]\d+)?)$`,
+	"i",
+);
+// Schemes whose host grammar is identifier-shaped, so any trailing
+// `:<selector-chunk>` is unambiguously a read-tool selector. `mcp://` is
+// excluded because mcp resource URIs may legitimately contain colons. `ssh://`
+// is included despite an optional `:port`; `splitInternalUrlSel` skips the peel
+// for an `ssh://host:port` that has no `/path`, so the port colon is never
+// mistaken for a selector (a real ssh selector trails the `/path`, e.g.
+// `ssh://h/f:1-5`).
+const INTERNAL_SCHEMES_WITH_SELECTORS: Record<string, true> = {
+	agent: true,
+	artifact: true,
+	issue: true,
+	local: true,
+	memory: true,
+	omp: true,
+	pr: true,
+	rule: true,
+	skill: true,
+	ssh: true,
+	vault: true,
+};
+// Schemes whose resource URIs are server-defined and may legitimately end
+// with selector-shaped tails (e.g. `:raw`, `:conflicts`, `:1-50`, `/:raw`).
+// `McpProtocolHandler` resolves by exact URI match (`r.uri === uri`), so
+// peeling syntactically can make valid resources unreachable. Keep these
+// schemes opaque; selector support for them needs a resolver-aware path that
+// tries the exact URI before interpreting any suffix as a read selector.
+const OPAQUE_RESOURCE_SCHEMES: ReadonlySet<string> = new Set(["mcp"]);
+const INTERNAL_URL_SCHEME_RE = /^([a-z][a-z0-9+.-]*):\/\//i;
+const NARROW_NO_BREAK_SPACE = "\u202F";
+const TOP_LEVEL_INTERNAL_URL_PREFIXES = [
+	"agent://",
+	"artifact://",
+	"skill://",
+	"rule://",
+	"local://",
+	"mcp://",
+	"ssh://",
+	"vault://",
+] as const;
+
+function normalizeUnicodeSpaces(str: string): string {
+	return str.replace(UNICODE_SPACES, " ");
+}
+
+function tryMacOSScreenshotPath(filePath: string): string {
+	return filePath.replace(/ (AM|PM)\./g, `${NARROW_NO_BREAK_SPACE}$1.`);
+}
+
+function tryNFDVariant(filePath: string): string {
+	// macOS stores filenames in NFD (decomposed) form, try converting user input to NFD
+	return filePath.normalize("NFD");
+}
+
+function tryCurlyQuoteVariant(filePath: string): string {
+	// macOS uses U+2019 (right single quotation mark) in screenshot names like "Capture d'écran"
+	// Users typically type U+0027 (straight apostrophe)
+	return filePath.replace(/'/g, "\u2019");
+}
+
+function tryShellEscapedPath(filePath: string): string {
+	if (!filePath.includes("\\") || !filePath.includes("/")) return filePath;
+	return filePath.replace(/\\([ \t"'(){}[\]])/g, "$1");
+}
+
+function fileExists(filePath: string): boolean {
+	try {
+		fs.accessSync(filePath, fs.constants.F_OK);
+		return true;
+	} catch {
+		return false;
+	}
+}
+
+function normalizeAtPrefix(filePath: string): string {
+	if (!filePath.startsWith("@")) return filePath;
+
+	const withoutAt = filePath.slice(1);
+
+	// We only treat a leading "@" as a shorthand for a small set of well-known
+	// syntaxes. This avoids mangling literal paths like "@my-file.txt".
+	if (
+		withoutAt.startsWith("/") ||
+		withoutAt === "~" ||
+		withoutAt.startsWith("~/") ||
+		// Windows absolute paths (drive letters / UNC / root-relative)
+		path.win32.isAbsolute(withoutAt) ||
+		// Internal URL shorthands
+		withoutAt.startsWith("agent://") ||
+		withoutAt.startsWith("artifact://") ||
+		withoutAt.startsWith("skill://") ||
+		withoutAt.startsWith("rule://") ||
+		withoutAt.startsWith("local:") ||
+		withoutAt.startsWith("mcp://")
+	) {
+		return withoutAt;
+	}
+
+	return filePath;
+}
+
+function stripFileUrl(filePath: string): string {
+	if (!filePath.toLowerCase().startsWith("file://")) return filePath;
+
+	try {
+		return url.fileURLToPath(filePath);
+	} catch {
+		return filePath;
+	}
+}
+
+export function expandTilde(filePath: string, home?: string): string {
+	const h = home ?? os.homedir();
+	if (filePath === "~") return h;
+	if (filePath.startsWith("~/") || filePath.startsWith("~\\")) {
+		return h + filePath.slice(1);
+	}
+	if (filePath.startsWith("~")) {
+		return path.join(h, filePath.slice(1));
+	}
+	return filePath;
+}
+
+export function expandPath(filePath: string): string {
+	const normalized = stripWindowsExtendedLengthPathPrefix(
+		stripFileUrl(normalizeUnicodeSpaces(normalizeAtPrefix(filePath))),
+	);
+	return expandTilde(normalized);
+}
+
+function isAsciiDriveLetter(value: string): boolean {
+	if (value.length !== 1) return false;
+	const code = value.charCodeAt(0);
+	return (code >= 65 && code <= 90) || (code >= 97 && code <= 122);
+}
+
+function windowsDriveAliasPath(filePath: string): string | undefined {
+	if (!filePath.startsWith("/")) return undefined;
+	const parts = filePath.split("/");
+	if (parts[0] !== "") return undefined;
+
+	let drive: string | undefined;
+	let tailStart = 2;
+	if (parts.length >= 2 && isAsciiDriveLetter(parts[1] ?? "")) {
+		drive = parts[1]!.toUpperCase();
+	} else if (parts.length >= 3 && (parts[1] ?? "").toLowerCase() === "mnt" && isAsciiDriveLetter(parts[2] ?? "")) {
+		drive = parts[2]!.toUpperCase();
+		tailStart = 3;
+	}
+	if (!drive) return undefined;
+
+	const tail = parts.slice(tailStart).filter(Boolean).join("\\");
+	return tail ? `${drive}:\\${tail}` : `${drive}:\\`;
+}
+
+export function normalizeWindowsDriveAliasPath(filePath: string, platform: NodeJS.Platform = process.platform): string {
+	if (platform !== "win32") return filePath;
+	return windowsDriveAliasPath(filePath) ?? filePath;
+}
+
+/**
+ * Inclusive line range describing one selector segment (e.g. `50-100`,
+ * `301-`, or `50+10`). `endLine` is `undefined` for open-ended ranges.
+ */
+export interface LineRange {
+	startLine: number;
+	endLine: number | undefined;
+}
+
+const LINE_RANGE_CHUNK_RE = /^L?(\d+)(?:(\.\.|[-+])L?(\d+)?)?$/i;
+
+/** Parse a single `N`, `N-M`, `N-`, `N+K`, or `..`-aliased (`N..M`, `N..`) chunk. Throws via {@link ToolError} on invalid bounds. */
+export function parseLineRangeChunk(sel: string): LineRange | null {
+	const lineMatch = LINE_RANGE_CHUNK_RE.exec(sel);
+	if (!lineMatch) return null;
+	const rawStart = Number.parseInt(lineMatch[1]!, 10);
+	if (rawStart < 1) {
+		throw new ToolError("Line selector 0 is invalid; lines are 1-indexed. Use :1.");
+	}
+	// `..` is a forgiving alias for `-` (e.g. `2724..2727` == `2724-2727`).
+	const sep = lineMatch[2] === ".." ? "-" : lineMatch[2];
+	const rhs = lineMatch[3] ? Number.parseInt(lineMatch[3], 10) : undefined;
+	let rawEnd: number | undefined;
+	if (sep === "+") {
+		if (rhs === undefined || rhs < 1) {
+			throw new ToolError(`Invalid range ${rawStart}+${rhs ?? 0}: count must be >= 1.`);
+		}
+		rawEnd = rawStart + rhs - 1;
+	} else if (sep === "-") {
+		// `301-` is shorthand for "from 301 onward" — equivalent to bare `301`.
+		if (rhs !== undefined) {
+			if (rhs < rawStart) {
+				throw new ToolError(`Invalid range ${rawStart}-${rhs}: end must be >= start.`);
+			}
+			rawEnd = rhs;
+		}
+	}
+	return { startLine: rawStart, endLine: rawEnd };
+}
+
+/**
+ * Parse a comma-separated list of line ranges (e.g. `5-16,960-973`). Returns
+ * the ranges in ascending order with overlapping/adjacent ranges merged so
+ * downstream consumers can stream the file in a single forward pass per range.
+ */
+export function parseLineRanges(sel: string): [LineRange, ...LineRange[]] | null {
+	const chunks = sel.split(",");
+	const parsed: LineRange[] = [];
+	for (const chunk of chunks) {
+		const range = parseLineRangeChunk(chunk);
+		if (!range) return null;
+		parsed.push(range);
+	}
+	if (parsed.length === 0) return null;
+	parsed.sort((a, b) => a.startLine - b.startLine);
+
+	const merged: LineRange[] = [parsed[0]];
+	for (let i = 1; i < parsed.length; i++) {
+		const current = parsed[i];
+		const last = merged[merged.length - 1];
+		// Open-ended (endLine undefined) means "to EOF" — any later range is absorbed.
+		if (last.endLine === undefined) continue;
+		// Merge when current starts within (or immediately after) the last range.
+		if (current.startLine <= last.endLine + 1) {
+			if (current.endLine === undefined || current.endLine > last.endLine) {
+				merged[merged.length - 1] = { startLine: last.startLine, endLine: current.endLine };
+			}
+			continue;
+		}
+		merged.push(current);
+	}
+	return merged as [LineRange, ...LineRange[]];
+}
+
+/**
+ * Extract the line-range component from a read-tool selector that may also
+ * carry a verbatim/index display mode (`raw`, `conflicts`) — alone or compounded
+ * with a range (`raw:50-100`, `50-100:raw`). Returns the parsed ranges when the
+ * selector names any, otherwise `undefined` (pure `raw`/`conflicts`/none).
+ *
+ * Used by content search, which honors line ranges as a match filter but has no
+ * use for verbatim/conflict display modes — so those selectors are accepted and
+ * treated as an unfiltered, whole-resource search rather than rejected.
+ */
+export function selectorLineRanges(sel: string | undefined): [LineRange, ...LineRange[]] | undefined {
+	if (!sel) return undefined;
+	for (const chunk of sel.split(":")) {
+		const lower = chunk.toLowerCase();
+		if (lower === "raw" || lower === "conflicts") continue;
+		const ranges = parseLineRanges(chunk);
+		if (ranges) return ranges;
+	}
+	return undefined;
+}
+
+/** Return `true` when `lineNumber` (1-indexed) falls in any of the supplied ranges. */
+export function isLineInRanges(lineNumber: number, ranges: readonly LineRange[]): boolean {
+	for (const range of ranges) {
+		if (lineNumber < range.startLine) continue;
+		if (range.endLine === undefined || lineNumber <= range.endLine) return true;
+	}
+	return false;
+}
+
+export function splitPathAndSel(rawPath: string): { path: string; sel?: string } {
+	const colon = rawPath.lastIndexOf(":");
+	if (colon <= 0) return { path: rawPath };
+
+	const candidate = rawPath.slice(colon + 1);
+	if (!FILE_LINE_RANGE_RE.test(candidate)) return { path: rawPath };
+
+	let basePath = rawPath.slice(0, colon);
+	let sel = candidate;
+
+	// Allow a compound trailing selector: `path:1-50:raw` or `path:raw:1-50`.
+	// The two chunks must be one line-range plus one `raw`, in either order.
+	const innerColon = basePath.lastIndexOf(":");
+	if (innerColon > 0) {
+		const innerCandidate = basePath.slice(innerColon + 1);
+		const innerIsRaw = FILE_RAW_ONLY_RE.test(innerCandidate);
+		const outerIsRaw = FILE_RAW_ONLY_RE.test(candidate);
+		const innerIsRange = FILE_LINE_RANGE_ONLY_RE.test(innerCandidate);
+		const outerIsRange = FILE_LINE_RANGE_ONLY_RE.test(candidate);
+		if ((innerIsRaw && outerIsRange) || (innerIsRange && outerIsRaw)) {
+			sel = `${innerCandidate}:${candidate}`;
+			basePath = basePath.slice(0, innerColon);
+		}
+	}
+
+	return { path: basePath, sel };
+}
+
+/**
+ * Variant of {@link splitPathAndSel} for internal URLs (`scheme://...`).
+ *
+ * The filesystem-path splitter is intentionally conservative: it refuses to
+ * peel a trailing `:<chunk>` unless that chunk matches the strict selector
+ * grammar. That rule is right for filesystem paths (a file named `a:1-50` is
+ * legal) but wrong for internal URLs, where any trailing `:<chunk>` after the
+ * scheme is unambiguously a read-tool selector — even if malformed (e.g.
+ * `artifact://3:raw:-100`).
+ *
+ * This function iteratively peels selector-shaped chunks (well-formed plus
+ * common malformed shapes like `:-N`) so the rest of the read tool can pass a
+ * clean URL to the protocol handler and surface selector errors via parseSel
+ * instead of as misleading "host invalid" errors from the handler. Schemes
+ * whose resource URIs may legitimately contain colons (`mcp://`) are skipped.
+ *
+ * Falls back to the input unchanged when nothing matches.
+ */
+
+export function splitInternalUrlSel(rawPath: string): { path: string; sel?: string } {
+	const schemeMatch = rawPath.match(INTERNAL_URL_SCHEME_RE);
+	if (!schemeMatch) return { path: rawPath };
+	const scheme = schemeMatch[1].toLowerCase();
+	// Opaque schemes (mcp://, etc.) carry server-defined resource URIs that may
+	// legitimately end in selector-shaped tails. Forward verbatim — see
+	// OPAQUE_RESOURCE_SCHEMES.
+	if (OPAQUE_RESOURCE_SCHEMES.has(scheme)) return { path: rawPath };
+	if (!INTERNAL_SCHEMES_WITH_SELECTORS[scheme]) return { path: rawPath };
+
+	const schemeEnd = schemeMatch[0].length;
+	// ssh:// authority carries an optional `:port`; with no `/path` after the
+	// authority, a trailing `:NNNN` is the port, not a read selector
+	// (e.g. ssh://host:2222). Other schemes' authority-trailing selectors
+	// (artifact://5:1-50) still peel, so this guard is ssh-specific.
+	if (scheme === "ssh" && rawPath.indexOf("/", schemeEnd) === -1) {
+		return { path: rawPath };
+	}
+	let path = rawPath;
+	const chunks: string[] = [];
+	while (true) {
+		const colon = path.lastIndexOf(":");
+		// Stop before crossing into the scheme separator `://`.
+		if (colon < schemeEnd) break;
+		const tail = path.slice(colon + 1);
+		if (!INTERNAL_URL_SELECTOR_PART_RE.test(tail)) break;
+		chunks.unshift(tail);
+		path = path.slice(0, colon);
+	}
+	if (chunks.length === 0) return { path: rawPath };
+	return { path, sel: chunks.join(":") };
+}
+
+/**
+ * Peel a read-tool selector off an internal-URL write target so `write` resolves
+ * the same file `read` does (e.g. `ssh://h/f:raw` -> `ssh://h/f`). Only the
+ * whole-file display modes `raw`/`conflicts` are accepted (they do not change
+ * which bytes are written); any other selector-shaped tail `splitInternalUrlSel`
+ * peels — a line range, a compound like `raw:1-20`, or a malformed `:-N` — throws,
+ * because `write` addresses a whole file, not a partial range, and silently
+ * stripping it would write to a path the caller never named. Non-URL paths and
+ * URLs without a selector pass through unchanged.
+ */
+export function peelWriteUrlSelector(rawPath: string): string {
+	const { path, sel } = splitInternalUrlSel(rawPath);
+	if (sel === undefined) return rawPath;
+	// Case-insensitive to match read's selector grammar (parseSel + the /i regexes above).
+	if (/^(?:raw|conflicts)$/i.test(sel)) return path;
+	throw new ToolError(
+		`write does not accept the trailing selector ":${sel}" — it writes a whole file. ` +
+			`Remove ":${sel}", or if the filename truly ends with it, percent-encode the ":" as %3A.`,
+	);
+}
+
+function assertNotInternalUrl(expanded: string, original: string): void {
+	for (const prefix of TOP_LEVEL_INTERNAL_URL_PREFIXES) {
+		if (expanded.startsWith(prefix)) {
+			throw new Error(
+				`Path "${original}" uses internal scheme "${prefix}" and must be resolved through the proper protocol handler, not as a filesystem path.`,
+			);
+		}
+	}
+}
+
+export function normalizeLocalScheme(filePath: string): string {
+	return filePath.replace(/^(local:)\/(?!\/)/, "$1//");
+}
+
+export function isInternalUrlPath(filePath: string): boolean {
+	const normalized = normalizeLocalScheme(filePath);
+	const expandedAndNormalized = normalizeLocalScheme(expandPath(normalized));
+	for (const prefix of TOP_LEVEL_INTERNAL_URL_PREFIXES) {
+		if (expandedAndNormalized.startsWith(prefix)) return true;
+	}
+	return false;
+}
+
+/**
+ * True when a tool path argument references the `ssh://` scheme anywhere.
+ *
+ * Substring (not anchored) on purpose: it feeds the read/search/write approval
+ * tier, which runs synchronously on the raw args. `search` only flattens a
+ * delimited `paths: "a,ssh://h/x"` into separate entries *after* approval, so an
+ * anchored check would let an embedded `ssh://` slip through at the read tier.
+ * Matching the literal `ssh://` substring also tracks exactly what routes to the
+ * SSH handler; over-matching only over-prompts (fail-closed).
+ */
+export function pathTargetsSsh(path: string): boolean {
+	return /ssh:\/\//i.test(path);
+}
+
+/**
+ * True when a path is specifically an `ssh://` URL (anchored scheme match).
+ * Unlike {@link pathTargetsSsh} (substring, for the pre-expansion approval
+ * scan), this is the exact per-entry check used to reject `ssh://` *before* a
+ * side-effecting `InternalUrlRouter.resolve` in tools that need a local file.
+ */
+export function isSshUrl(path: string): boolean {
+	return /^ssh:\/\//i.test(path.trim());
+}
+
+/**
+ * True when the read tool's URL parser (`parseReadUrlTarget` in fetch.ts) would
+ * recognize this path as a readable external URL: a strict `http(s)://`, a
+ * collapsed `http(s):/host` (Node path normalization folds `//` → `/`), or a
+ * scheme-less `www.` spelling. Keep in sync with `parseReadUrlTarget`.
+ */
+export function isReadableUrlPath(value: string): boolean {
+	return /^https?:\/\/?/i.test(value) || /^www\./i.test(value);
+}
+
+/**
+ * Resolve a path relative to the given cwd.
+ * Handles ~ expansion and absolute paths.
+ *
+ * A bare root slash is treated as a workspace-root alias for tool inputs. Users
+ * often pass `/` to mean “search from here”, and letting tools escape to the
+ * filesystem root is almost never what they intended.
+ */
+export function resolveToCwd(filePath: string, cwd: string): string {
+	const normalized = normalizeLocalScheme(filePath);
+	const expanded = normalizeWindowsDriveAliasPath(expandPath(normalized));
+	const expandedAndNormalized = normalizeLocalScheme(expanded);
+
+	assertNotInternalUrl(expandedAndNormalized, normalized);
+
+	if (/^\/+$/.test(expanded)) {
+		return cwd;
+	}
+	if (path.isAbsolute(expanded)) {
+		return expanded;
+	}
+	return path.resolve(cwd, expanded);
+}
+
+export function formatPathRelativeToCwd(
+	filePath: string,
+	cwd: string,
+	options: { trailingSlash?: boolean } = {},
+): string {
+	const resolvedCwd = path.resolve(cwd);
+	const normalized = normalizeLocalScheme(filePath);
+	if (isInternalUrlPath(normalized)) {
+		return normalized;
+	}
+	const expanded = expandPath(normalized);
+	const resolvedPath = path.isAbsolute(expanded) ? path.resolve(expanded) : path.resolve(cwd, expanded);
+	const relative = path.relative(resolvedCwd, resolvedPath);
+	const isWithinCwd = relative === "" || (!relative.startsWith("..") && !path.isAbsolute(relative));
+	let displayPath = normalizePosixPath(isWithinCwd ? relative || "." : resolvedPath);
+	if (options.trailingSlash && displayPath !== "." && !displayPath.endsWith("/")) {
+		displayPath += "/";
+	}
+	return displayPath;
+}
+
+/**
+ * Strip matching surrounding double quotes from a path string.
+ * Common when users paste quoted paths from Windows Explorer or shell copy-paste.
+ * Only double quotes — single quotes are valid POSIX filename characters.
+ * Tradeoff: a POSIX path literally starting AND ending with " would also be unquoted.
+ * Accepted because such names are virtually nonexistent in practice.
+ */
+export function stripOuterDoubleQuotes(input: string): string {
+	return input.startsWith('"') && input.endsWith('"') && input.length > 1 ? input.slice(1, -1) : input;
+}
+function normalizePathSeparators(input: string): string {
+	if (isInternalUrlPath(input)) return input;
+	if (!input.includes("\\")) return input;
+	return input.replace(/\\/g, "/");
+}
+
+export function normalizePathLikeInput(input: string): string {
+	return stripOuterDoubleQuotes(input.trim());
+}
+
+/**
+ * Parse a JSON-encoded array of path strings (e.g. `'["a.ts","b.ts"]'`).
+ * Returns `null` when the input is not a bracketed JSON string array, so the
+ * caller can fall back to treating the input as a single literal path.
+ */
+function parseStringEncodedPathArray(input: string): string[] | null {
+	const trimmed = input.trim();
+	if (!trimmed.startsWith("[") || !trimmed.endsWith("]")) return null;
+
+	let parsed: unknown;
+	try {
+		parsed = JSON.parse(trimmed);
+	} catch {
+		return null;
+	}
+
+	if (!Array.isArray(parsed) || parsed.some(entry => typeof entry !== "string")) {
+		return null;
+	}
+	return parsed;
+}
+
+/**
+ * Normalize a path argument that may arrive as a single string, a JSON-encoded
+ * string array (`'["a.ts"]'`), or an actual array into a flat `string[]`.
+ * Delimited single strings (`"a.ts b.ts"`) are left for
+ * {@link expandDelimitedPathEntries} to split.
+ */
+export function toPathList(input: string | string[] | undefined): string[] {
+	if (typeof input === "string") return parseStringEncodedPathArray(input) ?? [input];
+	return input ?? [];
+}
+
+const GLOB_PATH_CHARS = ["*", "?", "[", "{"] as const;
+
+export function hasGlobPathChars(filePath: string): boolean {
+	return GLOB_PATH_CHARS.some(char => filePath.includes(char));
+}
+
+type PathEntrySplitter = (item: string) => { basePath: string };
+
+const TOP_LEVEL_WHITESPACE_RE = /\s/;
+
+type DelimitedPathSplitMode = "comma" | "semicolon" | "whitespace" | "mixed";
+
+function isDelimitedPathSeparator(ch: string, mode: DelimitedPathSplitMode): boolean {
+	if (mode === "comma") return ch === ",";
+	if (mode === "semicolon") return ch === ";";
+	if (mode === "whitespace") return TOP_LEVEL_WHITESPACE_RE.test(ch);
+	return ch === "," || ch === ";" || TOP_LEVEL_WHITESPACE_RE.test(ch);
+}
+
+function hasTopLevelPathDelimiter(entry: string): boolean {
+	let braceDepth = 0;
+	for (let i = 0; i < entry.length; i++) {
+		const ch = entry[i];
+		if (ch === "\\" && i + 1 < entry.length) {
+			i++;
+			continue;
+		}
+		if (ch === "{") {
+			braceDepth++;
+			continue;
+		}
+		if (ch === "}") {
+			if (braceDepth > 0) braceDepth--;
+			continue;
+		}
+		if (braceDepth === 0 && (ch === "," || ch === ";" || TOP_LEVEL_WHITESPACE_RE.test(ch))) {
+			return true;
+		}
+	}
+	return false;
+}
+
+function splitTopLevelDelimitedPath(entry: string, mode: DelimitedPathSplitMode): string[] {
+	const parts: string[] = [];
+	let braceDepth = 0;
+	let start = 0;
+	for (let i = 0; i < entry.length; i++) {
+		const ch = entry[i];
+		if (ch === "\\" && i + 1 < entry.length) {
+			i++;
+			continue;
+		}
+		if (ch === "{") {
+			braceDepth++;
+			continue;
+		}
+		if (ch === "}") {
+			if (braceDepth > 0) braceDepth--;
+			continue;
+		}
+		if (braceDepth !== 0 || !isDelimitedPathSeparator(ch, mode)) continue;
+		parts.push(entry.slice(start, i));
+		start = i + 1;
+	}
+	parts.push(entry.slice(start));
+	return parts;
+}
+
+async function delimitedPathPartResolves(entry: string, cwd: string, splitter: PathEntrySplitter): Promise<boolean> {
+	if (isInternalUrlPath(entry)) return true;
+	const peeled = splitPathAndSel(entry).path;
+	const { basePath } = splitter(peeled);
+	const absoluteBasePath = resolveToCwd(basePath, cwd);
+	try {
+		await fs.promises.stat(absoluteBasePath);
+		return true;
+	} catch (err) {
+		if (isEnoent(err)) return false;
+		throw err;
+	}
+}
+
+/**
+ * How many split parts must resolve to an existing path for the split to win.
+ * Semicolon is the documented list delimiter, so it splits unconditionally
+ * (`"none"`) — an all-missing list must still fan out so multi-path missing
+ * semantics can name every entry. Comma is legacy recovery (`"some"`), and
+ * whitespace/mixed are aggressive heuristics gated on every part existing.
+ */
+type DelimitedResolveRequirement = "all" | "some" | "none";
+
+async function tryDelimitedPathSplit(
+	entry: string,
+	cwd: string,
+	splitter: PathEntrySplitter,
+	mode: DelimitedPathSplitMode,
+	requirement: DelimitedResolveRequirement,
+): Promise<string[] | null> {
+	const rawParts = splitTopLevelDelimitedPath(entry, mode);
+	if (rawParts.length < 2) return null;
+
+	const parts = rawParts.map(normalizePathLikeInput).filter(part => part.length > 0);
+	if (parts.length === 0) return null;
+	if (parts.length < 2 && rawParts.length === parts.length) return null;
+
+	if (requirement !== "none") {
+		const resolved = await Promise.all(parts.map(part => delimitedPathPartResolves(part, cwd, splitter)));
+		const valid = requirement === "all" ? resolved.every(Boolean) : resolved.some(Boolean);
+		if (!valid) return null;
+	}
+	return parts;
+}
+
+/**
+ * Split one path-like entry whose multiple targets were flattened into one
+ * string. Existing paths are kept intact, so real filenames containing spaces,
+ * commas, or semicolons win over delimiter recovery.
+ */
+export async function splitDelimitedPathEntry(
+	entry: string,
+	cwd: string,
+	options: { splitter?: PathEntrySplitter } = {},
+): Promise<string[] | null> {
+	const normalizedEntry = normalizePathLikeInput(entry);
+	if (!hasTopLevelPathDelimiter(normalizedEntry)) return null;
+	if (isInternalUrlPath(normalizedEntry)) return null;
+
+	const splitter = options.splitter ?? parseSearchPath;
+	const peeledEntry = splitPathAndSel(normalizedEntry).path;
+	if (!hasGlobPathChars(peeledEntry) && (await delimitedPathPartResolves(normalizedEntry, cwd, splitter))) {
+		return null;
+	}
+
+	return (
+		(await tryDelimitedPathSplit(normalizedEntry, cwd, splitter, "semicolon", "none")) ??
+		(await tryDelimitedPathSplit(normalizedEntry, cwd, splitter, "comma", "some")) ??
+		(await tryDelimitedPathSplit(normalizedEntry, cwd, splitter, "whitespace", "all")) ??
+		(await tryDelimitedPathSplit(normalizedEntry, cwd, splitter, "mixed", "all"))
+	);
+}
+
+/** Expand delimited entries in-place while preserving unsplit entries. */
+export async function expandDelimitedPathEntries(
+	entries: readonly string[],
+	cwd: string,
+	options: { splitter?: PathEntrySplitter } = {},
+): Promise<string[]> {
+	const expanded: string[] = [];
+	for (const entry of entries) {
+		const normalizedEntry = normalizePathLikeInput(entry);
+		const split = await splitDelimitedPathEntry(normalizedEntry, cwd, options);
+		if (split) expanded.push(...split);
+		else expanded.push(normalizedEntry);
+	}
+	return expanded;
+}
+
+export interface ParsedSearchPath {
+	basePath: string;
+	glob?: string;
+}
+
+export interface ParsedFindPattern {
+	basePath: string;
+	globPattern: string;
+	hasGlob: boolean;
+}
+
+export interface ResolvedSearchTarget {
+	basePath: string;
+	glob?: string;
+}
+
+export interface ResolvedMultiSearchPath {
+	basePath: string;
+	glob?: string;
+	scopePath: string;
+	exactFilePaths?: string[];
+	targets?: ResolvedSearchTarget[];
+}
+
+export interface ResolvedFindTarget {
+	basePath: string;
+	globPattern: string;
+	hasGlob: boolean;
+}
+
+export interface ResolvedMultiFindPattern {
+	targets: ResolvedFindTarget[];
+	scopePath: string;
+}
+export function parseSearchPath(filePath: string): ParsedSearchPath {
+	const normalizedPath = normalizePathSeparators(filePath);
+	const segments = normalizedPath.split("/");
+	let firstGlobIndex = -1;
+	for (let i = 0; i < segments.length; i++) {
+		if (hasGlobPathChars(segments[i])) {
+			firstGlobIndex = i;
+			break;
+		}
+	}
+
+	if (firstGlobIndex === -1) {
+		return { basePath: normalizedPath };
+	}
+
+	if (firstGlobIndex <= 0) {
+		return { basePath: ".", glob: normalizedPath };
+	}
+
+	return {
+		basePath: segments.slice(0, firstGlobIndex).join("/"),
+		glob: segments.slice(firstGlobIndex).join("/"),
+	};
+}
+
+/**
+ * Async sibling of {@link parseSearchPath} that prefers literal interpretation
+ * when a path containing glob metacharacters resolves to an existing entry on
+ * disk. Disambiguates Next.js/SvelteKit routes like `apps/[id]/page.tsx` —
+ * without this, `[id]` is parsed as a glob character class and silently
+ * matches nothing.
+ */
+export async function parseSearchPathPreferringLiteral(filePath: string, cwd: string): Promise<ParsedSearchPath> {
+	if (!hasGlobPathChars(filePath) || isInternalUrlPath(filePath)) return parseSearchPath(filePath);
+	try {
+		await fs.promises.stat(resolveToCwd(filePath, cwd));
+		return { basePath: normalizePathSeparators(filePath) };
+	} catch {
+		return parseSearchPath(filePath);
+	}
+}
+
+// Parse a find pattern into a base directory path and a glob pattern.
+// Examples:
+//   src/app/**/\*.tsx -> { basePath: "src/app", globPattern: "**/*.tsx", hasGlob: true }
+//   src/app/\*.tsx -> { basePath: "src/app", globPattern: "*.tsx", hasGlob: true }
+//   \*.ts -> { basePath: ".", globPattern: "**/*.ts", hasGlob: true }
+//   **/\*.json -> { basePath: ".", globPattern: "**/*.json", hasGlob: true }
+//   /abs/path/**/\*.ts -> { basePath: "/abs/path", globPattern: "**/*.ts", hasGlob: true }
+//   src/app -> { basePath: "src/app", globPattern: "**/*", hasGlob: false }
+export function parseFindPattern(pattern: string): ParsedFindPattern {
+	const normalizedPattern = normalizePathSeparators(pattern);
+	const segments = normalizedPattern.split("/");
+	let firstGlobIndex = -1;
+	for (let i = 0; i < segments.length; i++) {
+		if (hasGlobPathChars(segments[i])) {
+			firstGlobIndex = i;
+			break;
+		}
+	}
+
+	if (firstGlobIndex === -1) {
+		return { basePath: normalizedPattern, globPattern: "**/*", hasGlob: false };
+	}
+
+	if (firstGlobIndex === 0) {
+		const needsRecursive = !normalizedPattern.startsWith("**/");
+		return {
+			basePath: ".",
+			globPattern: needsRecursive ? `**/${normalizedPattern}` : normalizedPattern,
+			hasGlob: true,
+		};
+	}
+
+	return {
+		basePath: segments.slice(0, firstGlobIndex).join("/"),
+		globPattern: segments.slice(firstGlobIndex).join("/"),
+		hasGlob: true,
+	};
+}
+
+export function combineSearchGlobs(prefixGlob?: string, suffixGlob?: string): string | undefined {
+	if (!prefixGlob) return suffixGlob;
+	if (!suffixGlob) return prefixGlob;
+
+	const normalizedPrefix = prefixGlob.replace(/\/+$/, "");
+	const normalizedSuffix = suffixGlob.replace(/^\/+/, "");
+
+	return `${normalizedPrefix}/${normalizedSuffix}`;
+}
+
+function normalizePosixPath(filePath: string): string {
+	return filePath.replace(/\\/g, "/");
+}
+
+function joinRelativeGlob(basePath: string | undefined, globPattern: string): string {
+	if (!basePath || basePath === ".") return normalizePosixPath(globPattern).replace(/^\/+/, "");
+	const normalizedBase = normalizePosixPath(basePath).replace(/\/+$/, "");
+	const normalizedGlob = normalizePosixPath(globPattern).replace(/^\/+/, "");
+	return `${normalizedBase}/${normalizedGlob}`;
+}
+
+function buildBraceUnion(patterns: string[]): string | undefined {
+	const uniquePatterns = [...new Set(patterns.map(pattern => normalizePosixPath(pattern).trim()).filter(Boolean))];
+	if (uniquePatterns.length === 0) return undefined;
+	if (uniquePatterns.length === 1) return uniquePatterns[0];
+	return `{${uniquePatterns.join(",")}}`;
+}
+
+function findCommonBasePath(paths: string[]): string {
+	if (paths.length === 0) return ".";
+	let commonParts = path.resolve(paths[0]).split(path.sep);
+	for (const candidatePath of paths.slice(1)) {
+		const candidateParts = path.resolve(candidatePath).split(path.sep);
+		let sharedCount = 0;
+		const maxShared = Math.min(commonParts.length, candidateParts.length);
+		while (sharedCount < maxShared && commonParts[sharedCount] === candidateParts[sharedCount]) {
+			sharedCount += 1;
+		}
+		commonParts = commonParts.slice(0, sharedCount);
+	}
+	if (commonParts.length === 0) {
+		return path.parse(path.resolve(paths[0])).root;
+	}
+	const joined = commonParts.join(path.sep);
+	return joined || path.parse(path.resolve(paths[0])).root;
+}
+
+function toScopeDisplay(items: string[], cwd: string): string {
+	return items
+		.map(item =>
+			formatPathRelativeToCwd(item, cwd, {
+				trailingSlash: item.endsWith("/") || item.endsWith("\\"),
+			}),
+		)
+		.join(", ");
+}
+
+async function resolveSearchPathItems(
+	pathItems: string[],
+	cwd: string,
+	suffixGlob?: string,
+	fanOutFileItems = false,
+): Promise<ResolvedMultiSearchPath | undefined> {
+	if (pathItems.length < 1) {
+		return undefined;
+	}
+
+	const parsedItems = await Promise.all(
+		pathItems.map(async item => {
+			const parsedPath = await parseSearchPathPreferringLiteral(item, cwd);
+			const absoluteBasePath = resolveToCwd(parsedPath.basePath, cwd);
+			const stat = await fs.promises.stat(absoluteBasePath);
+			return { raw: item, parsedPath, absoluteBasePath, stat };
+		}),
+	);
+
+	const allExactFiles = !suffixGlob && parsedItems.every(item => !item.parsedPath.glob && item.stat.isFile());
+	const commonBasePath = findCommonBasePath(parsedItems.map(item => item.absoluteBasePath));
+	const combinedPatterns = parsedItems.map(item => {
+		const relativeBasePath = normalizePosixPath(path.relative(commonBasePath, item.absoluteBasePath)) || ".";
+		if (item.parsedPath.glob) {
+			const pathGlob = joinRelativeGlob(relativeBasePath, item.parsedPath.glob);
+			return combineSearchGlobs(pathGlob, suffixGlob) ?? pathGlob;
+		}
+		if (suffixGlob) {
+			const pathPrefix = relativeBasePath === "." ? undefined : relativeBasePath;
+			return combineSearchGlobs(pathPrefix, suffixGlob) ?? suffixGlob;
+		}
+		if (item.stat.isDirectory()) {
+			return joinRelativeGlob(relativeBasePath, "**/*");
+		}
+		return relativeBasePath === "." ? path.basename(item.absoluteBasePath) : relativeBasePath;
+	});
+	// A single walk rooted at the common ancestor is only safe when that
+	// ancestor is itself one of the requested scopes (e.g. `.` + `src/foo.ts`):
+	// the walk then covers exactly what the caller asked for. When the common
+	// ancestor is an unrequested parent (`.` + `~/.gitconfig` → `$HOME`, or
+	// disjoint trees → `/`), a collapsed walk traverses every unrelated sibling
+	// under it — fan out into per-item targets so each scan stays bounded to a
+	// requested path.
+	const commonIsRequestedScope = parsedItems.some(item => item.absoluteBasePath === commonBasePath);
+	// Walkers prune `.git` unconditionally and honor gitignore, so a plain-file
+	// item folded into a directory walk's glob union (`.` + `.git/config`) can
+	// silently never match. Callers that dedupe overlapping results opt in via
+	// `fanOutFileItems` to get explicit file targets, which bypass the walker.
+	const demotesFileItem =
+		fanOutFileItems && !allExactFiles && parsedItems.some(item => !item.parsedPath.glob && item.stat.isFile());
+	const targets =
+		parsedItems.length > 1 && (!commonIsRequestedScope || demotesFileItem)
+			? parsedItems.map(item => ({
+					basePath: item.absoluteBasePath,
+					glob: item.parsedPath.glob ? combineSearchGlobs(item.parsedPath.glob, suffixGlob) : suffixGlob,
+				}))
+			: undefined;
+
+	return {
+		basePath: commonBasePath,
+		glob: buildBraceUnion(combinedPatterns),
+		scopePath: toScopeDisplay(pathItems, cwd),
+		exactFilePaths: allExactFiles ? parsedItems.map(item => item.absoluteBasePath) : undefined,
+		targets,
+	};
+}
+
+export async function resolveExplicitSearchPaths(
+	pathItems: string[],
+	cwd: string,
+	suffixGlob?: string,
+	fanOutFileItems = false,
+): Promise<ResolvedMultiSearchPath | undefined> {
+	return resolveSearchPathItems([...new Set(pathItems)], cwd, suffixGlob, fanOutFileItems);
+}
+
+async function resolveFindPatternItems(
+	patternItems: string[],
+	cwd: string,
+): Promise<ResolvedMultiFindPattern | undefined> {
+	if (patternItems.length <= 1) {
+		return undefined;
+	}
+
+	// Each path becomes its own walk root. Collapsing to a shared common ancestor
+	// (and filtering with a brace-union glob) would force the walker to traverse
+	// and stat every unrelated sibling under that ancestor — two paths under
+	// $HOME would scan all of $HOME. The find tool fans these targets out in
+	// parallel instead, so every scan stays bounded to exactly one requested path.
+	const targets = patternItems.map(item => {
+		const parsedPattern = parseFindPattern(item);
+		return {
+			basePath: resolveToCwd(parsedPattern.basePath, cwd),
+			globPattern: parsedPattern.globPattern,
+			hasGlob: parsedPattern.hasGlob,
+		};
+	});
+
+	return {
+		targets,
+		scopePath: toScopeDisplay(patternItems, cwd),
+	};
+}
+
+export async function resolveExplicitFindPatterns(
+	patternItems: string[],
+	cwd: string,
+): Promise<ResolvedMultiFindPattern | undefined> {
+	return resolveFindPatternItems([...new Set(patternItems)], cwd);
+}
+
+/**
+ * Result of partitioning a list of user-supplied paths/globs into entries whose
+ * base directory currently exists on disk versus those that do not.
+ *
+ * Used by multi-path tools (search, find, ast_grep, ast_edit) to tolerate one
+ * or more missing entries in a multi-path call: the surviving entries should
+ * still be searched, with the missing entries surfaced as a non-fatal warning.
+ */
+export interface PartitionedPaths {
+	/** Raw input strings whose resolved base path exists. */
+	valid: string[];
+	/** Raw input strings whose resolved base path is missing (ENOENT). */
+	missing: string[];
+}
+
+/**
+ * Stat each input's base path concurrently; return entries split by existence.
+ *
+ * `splitter` is expected to be {@link parseFindPattern} or
+ * {@link parseSearchPath}: both return a `basePath` field that this helper
+ * resolves against `cwd` and stats. ENOENT is the only swallowed error — every
+ * other stat failure (permission, IO, etc.) propagates so callers do not silently
+ * skip paths that exist but are unreadable.
+ *
+ * Order of `valid` and `missing` follows the input order, so callers can rely
+ * on `valid[0]` matching the first surviving user-supplied entry.
+ */
+export async function partitionExistingPaths(
+	items: string[],
+	cwd: string,
+	splitter: (item: string) => { basePath: string },
+): Promise<PartitionedPaths> {
+	const settled = await Promise.all(
+		items.map(async item => {
+			const { basePath } = splitter(item);
+			const absoluteBasePath = resolveToCwd(basePath, cwd);
+			try {
+				await fs.promises.stat(absoluteBasePath);
+				return { item, exists: true } as const;
+			} catch (err) {
+				if (isEnoent(err)) return { item, exists: false } as const;
+				throw err;
+			}
+		}),
+	);
+	const valid: string[] = [];
+	const missing: string[] = [];
+	for (const entry of settled) {
+		if (entry.exists) valid.push(entry.item);
+		else missing.push(entry.item);
+	}
+	return { valid, missing };
+}
+
+export function resolveReadPath(filePath: string, cwd: string): string {
+	const resolved = resolveToCwd(filePath, cwd);
+	const shellEscapedVariant = tryShellEscapedPath(resolved);
+	const baseCandidates = shellEscapedVariant !== resolved ? [resolved, shellEscapedVariant] : [resolved];
+
+	for (const baseCandidate of baseCandidates) {
+		if (fileExists(baseCandidate)) {
+			return baseCandidate;
+		}
+	}
+
+	for (const baseCandidate of baseCandidates) {
+		// Try macOS AM/PM variant (narrow no-break space before AM/PM)
+		const amPmVariant = tryMacOSScreenshotPath(baseCandidate);
+		if (amPmVariant !== baseCandidate && fileExists(amPmVariant)) {
+			return amPmVariant;
+		}
+
+		// Try NFD variant (macOS stores filenames in NFD form)
+		const nfdVariant = tryNFDVariant(baseCandidate);
+		if (nfdVariant !== baseCandidate && fileExists(nfdVariant)) {
+			return nfdVariant;
+		}
+
+		// Try curly quote variant (macOS uses U+2019 in screenshot names)
+		const curlyVariant = tryCurlyQuoteVariant(baseCandidate);
+		if (curlyVariant !== baseCandidate && fileExists(curlyVariant)) {
+			return curlyVariant;
+		}
+
+		// Try combined NFD + curly quote (for French macOS screenshots like "Capture d'écran")
+		const nfdCurlyVariant = tryCurlyQuoteVariant(nfdVariant);
+		if (nfdCurlyVariant !== baseCandidate && fileExists(nfdCurlyVariant)) {
+			return nfdCurlyVariant;
+		}
+	}
+
+	return resolved;
+}
+
+// =============================================================================
+// Tool-scope resolution (search/ast tools)
+// =============================================================================
+
+/** Local file materialized from a readable external URL for shared tool-scope resolution. */
+export interface ResolvedExternalSearchUrl {
+	/** Absolute or cwd-relative file path to search. */
+	sourcePath: string;
+	/** True when the materialized file must not mint editable anchors. */
+	immutable?: boolean;
+}
+
+export interface ToolScopeOptions {
+	rawPaths: string[];
+	cwd: string;
+	/** Verb used in the "Cannot {action} internal URL without a backing file: …" message. */
+	internalUrlAction: string;
+	/** Collect absolute paths flagged immutable by their internal-URL handler. */
+	trackImmutableSources?: boolean;
+	/** Honor `exactFilePaths` from {@link resolveExplicitSearchPaths} (search-only). */
+	surfaceExactFilePaths?: boolean;
+	/** Fan plain-file entries out into per-target scans instead of folding them
+	 * into a directory walk's glob union (search-only: the caller must dedupe
+	 * matches from overlapping targets). */
+	fanOutFileTargets?: boolean;
+	/** Extra hint appended to "Path not found" when stat fails and the user supplied multiple paths. */
+	multipathStatHint?: string;
+	/** Calling session's settings — forwarded to the internal-URL router so caller-aware handlers (issue://, pr://) honor it. */
+	settings?: unknown;
+	/** Caller's abort signal — forwarded to the internal-URL router. */
+	signal?: AbortSignal;
+	/** Calling session's `local://` root mapping — pins resolutions to the calling session. */
+	localProtocolOptions?: LocalProtocolOptions;
+	/** Calling session's loaded skills — lets skill:// resolve without process-global state. */
+	skills?: readonly Skill[];
+	/** Materialize readable external URLs to local text files before scope derivation. */
+	resolveExternalUrl?: (rawPath: string) => Promise<ResolvedExternalSearchUrl | undefined>;
+}
+
+export interface ToolScopeResolution {
+	searchPath: string;
+	scopePath: string;
+	globFilter: string | undefined;
+	isDirectory: boolean;
+	multiTargets?: ResolvedSearchTarget[];
+	exactFilePaths?: string[];
+	missingPaths: string[];
+	immutableSourcePaths: Set<string>;
+}
+
+/**
+ * Shared path-input pipeline for `search`, `ast_grep`, and `ast_edit`:
+ *  1. normalize + reject empty paths,
+ *  2. resolve internal URLs through {@link InternalUrlRouter} to backing files,
+ *  3. partition existing vs missing when multiple paths are supplied,
+ *  4. derive a single search base path / glob, or a multi-target list,
+ *  5. stat the resolved base path so callers can branch on directory vs file scope.
+ */
+export async function resolveToolSearchScope(opts: ToolScopeOptions): Promise<ToolScopeResolution> {
+	const { rawPaths: inputs, cwd, internalUrlAction } = opts;
+	const normalizedRawPaths = inputs.map(normalizePathLikeInput);
+	if (normalizedRawPaths.some(rawPath => rawPath.length === 0)) {
+		throw new ToolError("Search scope entries must be non-empty paths or globs");
+	}
+	const rawPaths = await expandDelimitedPathEntries(normalizedRawPaths, cwd);
+	if (rawPaths.some(rawPath => rawPath.length === 0)) {
+		throw new ToolError("Search scope entries must be non-empty paths or globs");
+	}
+	// Strict external-URL schemes. `file://` is intentionally absent: it has
+	// local-path semantics (expandPath strips it downstream), so it flows through
+	// the ordinary filesystem pipeline instead of the external-URL resolver.
+	const strictExternalUrlRe = /^(?:https?|ftp|ws|wss):\/\//i;
+	const internalRouter = InternalUrlRouter.instance();
+	const resolvedPathInputs: string[] = [];
+	const immutableSourcePaths = new Set<string>();
+	for (const rawPath of rawPaths) {
+		let externalUrl = strictExternalUrlRe.test(rawPath);
+		if (!externalUrl && isReadableUrlPath(rawPath) && !hasGlobPathChars(rawPath)) {
+			// Fuzzy spelling the read parser accepts (`www.host/…`, collapsed
+			// `https:/host/…`). An existing local path wins over URL
+			// interpretation so a directory literally named `www.foo` stays
+			// searchable; only a definitive ENOENT/ENOTDIR flips to URL handling
+			// (any other stat error means the path exists — let the local
+			// pipeline surface it).
+			try {
+				await fs.promises.stat(resolveToCwd(rawPath, cwd));
+			} catch (err) {
+				externalUrl = isEnoent(err) || isEnotdir(err);
+			}
+		}
+		if (externalUrl) {
+			const resolved = opts.resolveExternalUrl ? await opts.resolveExternalUrl(rawPath) : undefined;
+			if (resolved) {
+				resolvedPathInputs.push(resolved.sourcePath);
+				if (opts.trackImmutableSources && resolved.immutable) {
+					immutableSourcePaths.add(path.resolve(resolved.sourcePath));
+				}
+				continue;
+			}
+			// Resolver missing or declined (e.g. ftp/ws/wss): fail explicitly
+			// instead of letting the local-path fallthrough surface a confusing
+			// "Path not found" for a URL-shaped input.
+			throw new ToolError(
+				`Cannot ${internalUrlAction} external URL: ${rawPath}. Use \`read\` to fetch web content, then search the returned text.`,
+			);
+		}
+		if (!internalRouter.canHandle(rawPath)) {
+			resolvedPathInputs.push(rawPath);
+			continue;
+		}
+		if (isSshUrl(rawPath)) {
+			throw new ToolError(
+				`Cannot ${internalUrlAction} a remote ssh:// path (no local file): ${rawPath}. Use \`read ${rawPath}\` to view it, or the \`search\` tool to grep remote files.`,
+			);
+		}
+		if (hasGlobPathChars(rawPath)) {
+			throw new ToolError(`Glob patterns are not supported for internal URLs: ${rawPath}`);
+		}
+		const resource = await internalRouter.resolve(rawPath, {
+			cwd,
+			settings: opts.settings,
+			signal: opts.signal,
+			localProtocolOptions: opts.localProtocolOptions,
+			skills: opts.skills,
+			// Tool-scope resolution only needs `sourcePath`; skip content
+			// materialization so large artifacts (or any handler that separates
+			// path from content) stay searchable without OOM risk.
+			pathOnly: true,
+		});
+		if (!resource.sourcePath) {
+			throw new ToolError(`Cannot ${internalUrlAction} internal URL without a backing file: ${rawPath}`);
+		}
+		if (opts.trackImmutableSources && resource.immutable) {
+			immutableSourcePaths.add(path.resolve(resource.sourcePath));
+		}
+		resolvedPathInputs.push(resource.sourcePath);
+	}
+
+	let missingPaths: string[] = [];
+	let effectivePaths = resolvedPathInputs;
+	if (resolvedPathInputs.length > 1) {
+		const partition = await partitionExistingPaths(resolvedPathInputs, cwd, parseSearchPath);
+		if (partition.valid.length === 0) {
+			throw new ToolError(`Path not found: ${partition.missing.join(", ")}`);
+		}
+		effectivePaths = partition.valid;
+		missingPaths = partition.missing;
+	}
+
+	let searchPath: string;
+	let scopePath: string;
+	let globFilter: string | undefined;
+	let multiTargets: ResolvedSearchTarget[] | undefined;
+	let exactFilePaths: string[] | undefined;
+	if (effectivePaths.length === 1) {
+		const parsedPath = await parseSearchPathPreferringLiteral(effectivePaths[0] ?? ".", cwd);
+		searchPath = resolveToCwd(parsedPath.basePath, cwd);
+		globFilter = parsedPath.glob;
+		scopePath = formatPathRelativeToCwd(searchPath, cwd);
+	} else {
+		const multiSearchPath = await resolveExplicitSearchPaths(
+			effectivePaths,
+			cwd,
+			undefined,
+			opts.fanOutFileTargets === true,
+		);
+		if (!multiSearchPath) {
+			throw new ToolError("`paths` must contain at least one path or glob");
+		}
+		searchPath = multiSearchPath.basePath;
+		multiTargets = multiSearchPath.targets;
+		if (opts.surfaceExactFilePaths) {
+			exactFilePaths = multiSearchPath.exactFilePaths;
+			globFilter = exactFilePaths || multiTargets ? undefined : multiSearchPath.glob;
+		} else {
+			globFilter = multiTargets ? undefined : multiSearchPath.glob;
+		}
+		scopePath = multiSearchPath.scopePath;
+	}
+
+	let isDirectory: boolean;
+	try {
+		const stat = await Bun.file(searchPath).stat();
+		isDirectory = stat.isDirectory();
+	} catch {
+		const hint = opts.multipathStatHint && rawPaths.length > 1 ? opts.multipathStatHint : "";
+		throw new ToolError(`Path not found: ${scopePath}${hint}`);
+	}
+
+	return {
+		searchPath,
+		scopePath,
+		globFilter,
+		isDirectory,
+		multiTargets,
+		exactFilePaths,
+		missingPaths,
+		immutableSourcePaths,
+	};
+}