srclawyer 0.6.1 → 0.6.3

package/README.md

@@ -2,7 +2,7 @@
 
 Static analysis engine that derives privacy policy documentation directly from source code, and keeps it in sync as the code changes.
 
-> **Beta.** The CLI (`init`, `scan`), the Layer 1 static analysis engine, and markdown privacy policy generation all work today. JavaScript/TypeScript (Express, Next.js) is the most tested path; Python (Flask) is newer, with FastAPI and Django explicitly detected and flagged as unsupported rather than silently skipped — see [Supported languages and frameworks](#supported-languages-and-frameworks) and [Known gaps](#known-gaps) below. LLM-assisted reasoning for ambiguous findings (Layer 2) is available now under the bring-your-own-key tier, documented below as one optional step in `init` — a managed (hosted) tier is planned but not yet available.
+> **Beta.** The CLI (`init`, `scan`), the Layer 1 static analysis engine, and markdown privacy policy generation all work today. JavaScript/TypeScript (Express, Next.js) is the most tested path; Python (Flask) is newer, with FastAPI and Django explicitly detected and flagged as unsupported rather than silently skipped — see [Supported languages and frameworks](#supported-languages-and-frameworks) and [Known gaps](#known-gaps) below. Optional AI-assisted review of ambiguous findings (Layer 2) is available as an advanced option with your own Anthropic API key — see [AI-assisted review](#ai-assisted-review-advanced) below; a hosted tier that needs no key is planned but not yet available.
 >
 > Scanning is manual-only (`srclawyer scan`, run on demand) — this is the free/solo tier's permanent behavior, not a placeholder. Automatic CI/CD updating (a scan that runs itself on every PR, or on a schedule) isn't available yet; it's scoped for a future paid/team tier. See `PRODUCT_SPEC.md` Section 8.5 for the plan.
 
@@ -90,10 +90,41 @@ Every finding traces back to a file and line (`src/engine/types.ts#Finding`).
 - **Data sent to a third party via a raw HTTP call isn't attributed to that processor.** SDK-integration detection (above) is anchored to known package imports. A call like Python's `requests.post("https://api.some-provider.com/...", json={...})` or JS's `fetch("https://api.some-provider.com/...")` — bypassing a dedicated SDK entirely — is invisible to this detection today, in both languages, regardless of how sensitive the data it sends is.
 - **Python support is based on one real-world corpus repository so far** (a Flask app; see `test-fixtures/golden-corpus/`), not the broader validation JS detection has had. Expect rougher edges on Flask codebases with unusual patterns.
 - **OpenAPI/Swagger scanning reports every schema under `components.schemas`, including response-only ones** — a schema that's only ever returned by the server (never submitted by a client) gets identical treatment to a real request-input schema, since nothing currently checks whether a schema is actually reachable from an operation's `requestBody` or `parameters`. On a large, schema-heavy spec this can mean a long "Items Requiring Manual Review" list that includes fields the product never collects, only ever returns. Separately, a schema defined **inline** inside a `requestBody` (not as a named, `$ref`'d `components.schemas` entry) isn't read at all today, in either direction — only named component schemas are scanned.
+- **Flask support covers direct request reads and same-function aliases only.** It detects `request.form[...]`, `request.args.get(...)`, `request.json[...]` and the like, plus one assignment hop within a function (`data = request.get_json()`, then `data['x']`). Form classes — Flask-WTF/WTForms `FlaskForm` subclasses with `StringField`/`PasswordField` declarations read via `form.email.data` — are not detected, and neither are fields rendered by template helpers such as `{{ form.email() }}`. A Flask app that collects data through WTForms may therefore show fewer collected fields than it actually has.
+- **Server-side template syntax inside HTML is not analyzed.** Jinja, Django, ERB and similar delimiters (`{{`, `{%`, `<%`, `{#`) are not interpreted: an inline `<script>` that isn't valid JavaScript until the template engine runs is skipped (listed in `--json` as `template-syntax`, with a one-line note), and form fields that a template generates rather than spells out as `<input>` tags are not seen.
+
+## AI-assisted review (advanced)
+
+Off by default, and not part of `init`'s questions. `scan` already surfaces anything it can't confidently classify under "Items Requiring Manual Review"; with AI-assisted review enabled it first asks Anthropic, under **your own API key**, to classify each such item. For each item Anthropic receives: an opaque per-scan id, the item's data category, its short description (which names the field), and the one source line it was found on with secrets redacted — plus a fixed instruction prompt and the model name. **Not sent:** the file's path, its line number, or any surrounding code (the source line itself goes as written, apart from redacted secrets). Items the model declines stay in the manual-review list, and items involving children's data are never auto-resolved.
+
+Enable it when creating the config:
+
+```bash
+srclawyer init --enable-ai-review                              # asks which env var holds your key (default ANTHROPIC_API_KEY)
+srclawyer init --enable-ai-review --api-key-env-var MY_KEY     # same, non-interactive
+export ANTHROPIC_API_KEY=...                                   # or whichever name you chose
+srclawyer scan
+```
+
+or, for an existing config (`init` never overwrites one), add this to `.srclawyer.config.yml` by hand:
+
+```yaml
+tier: byok
+llm:
+  provider: anthropic
+  apiKeyEnvVar: ANTHROPIC_API_KEY
+```
+
+Only the variable's *name* is stored in the config — the key itself is never written to disk. When it's enabled, `scan` prints a notice that redacted snippets are being sent to Anthropic; `scan --no-llm` skips it for a single run. A hosted tier, where SrcLawyer supplies the model access and no key is needed, is planned but not yet available: a config with `tier: managed` gets a clear "not available yet" error rather than silently doing nothing.
 
 ## Network calls
 
-Detection itself is fully local and offline — no code leaves your machine for pattern matching. Once detection completes, `scan` makes one batched network call to grade the confidence of what it found (field names and derived yes/no signals only, never source code), and, for scans that use Zod schemas as request validators, one additional call sending a small redacted code fragment. This runs on every scan regardless of tier and is unrelated to the separate, optional LLM-assisted (Layer 2) reasoning described above. See `PRODUCT_SPEC.md`'s "Static Analysis Depth" section for the precise wording.
+Detection itself is fully local and offline — no code leaves your machine for pattern matching. Once detection completes, every `scan` makes these calls to a SrcLawyer-run service:
+
+- **Confidence grading** (one batched call). For each detected field: an opaque per-scan id (`f0`, `f1`, …), the data category its name maps to (e.g. `email`), and three derived signals — whether the name is ambiguous, whether it was read directly from the request in code (rather than inferred from a type), and how many confidently classified fields sit alongside it. **Not sent:** the field's name, file paths, line numbers, or source code.
+- **Zod-schema detection** (one more batched call, only for scans where Zod schemas appear to validate request data). Per schema: an opaque id (`c0`, `c1`, …) and a small, secret-redacted code fragment — the schema, the call that validates with it, and the imports it uses, as source text (so it includes the schema's field names and those import specifiers). **Not sent:** the schema's file path or line number.
+
+Ids are numbered per scan; the mapping back to files and lines never leaves your machine. Requests carry a shared API key (`x-api-key`) and the standard headers Node.js adds (such as `user-agent: node`); there is no install ID, machine information, or telemetry. These calls run on every scan, with or without the optional [AI-assisted review](#ai-assisted-review-advanced), and are separate from it. See `PRODUCT_SPEC.md`'s "Static Analysis Depth" section for the same wording.
 
 ## Secret handling
 

package/dist/cli.js

@@ -3,7 +3,7 @@ import { readFileSync } from "node:fs";
 import { dirname, join } from "node:path";
 import { fileURLToPath } from "node:url";
 import { Command } from "commander";
-import { runInit } from "./commands/init.js";
+import { isValidEnvVarName, runInit } from "./commands/init.js";
 import { runScan } from "./commands/scan.js";
 // Read from package.json rather than hardcoding, so --version can't silently drift out of sync with
 // the published version the way a literal string did through every prior release.
@@ -37,7 +37,15 @@ program
     .option("--minimum-age <number>", "minimum age required to use this product")
     .option("--sends-data-to-ai-services", "declare that user data is sent to third-party AI services")
     .option("--no-sends-data-to-ai-services", "declare that user data is not sent to third-party AI services")
+    .option("--enable-ai-review", "advanced: enable optional AI-assisted review of ambiguous findings with your own Anthropic API key (sets tier: byok)")
+    .option("--api-key-env-var <name>", "with --enable-ai-review: the environment variable holding your Anthropic API key (default ANTHROPIC_API_KEY)")
     .action(async (opts) => {
+    if (opts.apiKeyEnvVar !== undefined) {
+        if (!opts.enableAiReview)
+            program.error("error: --api-key-env-var requires --enable-ai-review");
+        if (!isValidEnvVarName(opts.apiKeyEnvVar))
+            program.error(`error: --api-key-env-var must be a valid environment variable name (got "${opts.apiKeyEnvVar}")`);
+    }
     let minimumAge;
     if (opts.minimumAge !== undefined) {
         const parsed = Number(opts.minimumAge);
@@ -60,6 +68,8 @@ program
         accountDeletionInstructions: opts.accountDeletionInstructions,
         minimumAge,
         sendsDataToAiServices: opts.sendsDataToAiServices,
+        enableAiReview: opts.enableAiReview,
+        apiKeyEnvVar: opts.apiKeyEnvVar,
     });
 });
 program

package/dist/cloud/protectedLogicClient.js

@@ -39,6 +39,22 @@ function chunkByCountAndBytes(items, maxCount, maxBytes) {
         batches.push(current);
     return batches;
 }
+/**
+ * The one place ids are assigned for anything sent to the Worker. Callers identify items by their own
+ * ids (finding ids embed file path, line and field name; Zod candidate ids ARE the file path), but
+ * those never leave this process: every item goes out under an opaque per-scan id (f0, f1, ... /
+ * c0, c1, ...), unique across all batches, and responses are mapped back here. The Worker only echoes
+ * ids (it never parses them), so nothing about the wire format depends on their shape.
+ */
+function toOpaque(items, prefix) {
+    const originalByOpaque = new Map();
+    const wire = items.map((item, i) => {
+        const opaque = `${prefix}${i}`;
+        originalByOpaque.set(opaque, item.id);
+        return { ...item, id: opaque };
+    });
+    return { wire, originalByOpaque };
+}
 function sleep(ms) {
     return new Promise((resolve) => setTimeout(resolve, ms));
 }
@@ -79,11 +95,15 @@ export async function scoreFields(items, options) {
     const batchMaxBytes = options.batchMaxBytes ?? DEFAULT_BATCH_MAX_BYTES;
     const timeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS;
     let error = null;
-    for (const batch of chunkByCountAndBytes(items, batchSize, batchMaxBytes)) {
+    const { wire, originalByOpaque } = toOpaque(items, "f");
+    for (const batch of chunkByCountAndBytes(wire, batchSize, batchMaxBytes)) {
         try {
             const response = await postJson(`${options.baseUrl}/v1/score-fields`, options.apiKey, { fields: batch }, timeoutMs);
-            for (const result of response.results)
-                results.set(result.id, result);
+            for (const result of response.results) {
+                const original = originalByOpaque.get(result.id);
+                if (original !== undefined)
+                    results.set(original, { ...result, id: original });
+            }
         }
         catch (err) {
             error = err instanceof Error ? err.message : String(err);
@@ -99,11 +119,15 @@ export async function detectZodSchemas(items, options) {
     const batchMaxBytes = options.batchMaxBytes ?? DEFAULT_BATCH_MAX_BYTES;
     const timeoutMs = options.timeoutMs ?? DEFAULT_TIMEOUT_MS;
     let error = null;
-    for (const batch of chunkByCountAndBytes(items, batchSize, batchMaxBytes)) {
+    const { wire, originalByOpaque } = toOpaque(items, "c");
+    for (const batch of chunkByCountAndBytes(wire, batchSize, batchMaxBytes)) {
         try {
             const response = await postJson(`${options.baseUrl}/v1/detect-zod-schema`, options.apiKey, { candidates: batch }, timeoutMs);
-            for (const result of response.results)
-                results.set(result.id, result);
+            for (const result of response.results) {
+                const original = originalByOpaque.get(result.id);
+                if (original !== undefined)
+                    results.set(original, { ...result, id: original });
+            }
         }
         catch (err) {
             error = err instanceof Error ? err.message : String(err);

package/dist/commands/init.d.ts

@@ -12,5 +12,10 @@ export interface InitFlags {
     accountDeletionInstructions?: string;
     minimumAge?: number;
     sendsDataToAiServices?: boolean;
+    /** Advanced, opt-in: sets tier: byok. Never asked about in the default flow. */
+    enableAiReview?: boolean;
+    apiKeyEnvVar?: string;
 }
+export declare const DEFAULT_API_KEY_ENV_VAR = "ANTHROPIC_API_KEY";
+export declare function isValidEnvVarName(name: string): boolean;
 export declare function runInit(root: string, flags?: InitFlags): Promise<void>;

package/dist/commands/init.js

@@ -2,7 +2,13 @@ import { existsSync, readFileSync } from "node:fs";
 import { join } from "node:path";
 import { createInterface } from "node:readline/promises";
 import { CONFIG_FILENAME, DEFAULT_CONFIG, configPath, hasUnmigratedLegacyConfig, legacyConfigMessage, writeConfig, } from "../config/config.js";
-import { NETWORK_CALLS_NOTICE } from "../policy/legalNotices.js";
+import { NETWORK_CALLS_NOTICE, byokNetworkNotice } from "../policy/legalNotices.js";
+import { runScan } from "./scan.js";
+export const DEFAULT_API_KEY_ENV_VAR = "ANTHROPIC_API_KEY";
+const ENV_VAR_NAME = /^[A-Za-z_][A-Za-z0-9_]*$/;
+export function isValidEnvVarName(name) {
+    return ENV_VAR_NAME.test(name);
+}
 function guessIndustry(root) {
     const pkgPath = join(root, "package.json");
     if (!existsSync(pkgPath))
@@ -82,13 +88,18 @@ function warnIfContradictsTargetMarkets(label, finalValue, suggestedFromMarkets)
 export async function runInit(root, flags = {}) {
     const existingPath = configPath(root);
     if (existsSync(existingPath)) {
-        console.log(`${CONFIG_FILENAME} already exists. Delete it first if you want to redo setup.`);
+        console.log(`${CONFIG_FILENAME} already exists -- left unchanged. To start over, delete it and re-run "srclawyer init":\n\n  rm ${CONFIG_FILENAME}\n  srclawyer init\n`);
         return;
     }
     if (hasUnmigratedLegacyConfig(root)) {
         console.log(legacyConfigMessage());
         return;
     }
+    // The final "run a scan now?" prompt below is interactive-only: skipped whenever this run is
+    // flag-driven (any flag was passed -- a scripted/CI invocation, even a partial one) or stdin isn't
+    // a real TTY (piped, redirected, non-interactive). Computed once, up front, rather than re-checked
+    // at the end, since nothing between here and there changes either condition.
+    const isInteractive = process.stdin.isTTY === true && Object.values(flags).every((v) => v === undefined);
     const guessedIndustry = guessIndustry(root);
     const rl = createInterface({ input: process.stdin, output: process.stdout });
     const lines = rl[Symbol.asyncIterator]();
@@ -126,19 +137,23 @@ export async function runInit(root, flags = {}) {
     // Deliberately about the SCANNED PRODUCT's own behavior (does it send its end-users' data to AI
     // services as part of its own functionality) -- not about SrcLawyer's own, separately-documented
     // BYOK network calls described in NETWORK_CALLS_NOTICE above.
-    const sendsDataToAiServices = await resolveField(flags.sendsDataToAiServices, gateYes, () => askYesNoOptional(lines, "Does this product send user data to third-party AI services (e.g. OpenAI, Anthropic) as part of its own functionality?"));
-    // "managed" is deliberately not offered here -- it's a real SrcLawyerTier value and config.ts/
-    // resolveProviderForTier.ts both still support it fully (a hand-edited config with tier: managed
-    // still gets a clear, accurate error at scan time), but there's no advertised path to it from this
-    // prompt, so a tester reading the choices here can't select a dead end during setup.
-    const enableByokRaw = await ask(lines, "Enable optional AI-assisted review with your own Anthropic key? (advanced)", "n");
-    const tier = /^y(es)?$/i.test(enableByokRaw.trim()) ? "byok" : "free";
+    const sendsDataToAiServices = await resolveField(flags.sendsDataToAiServices, gateYes, () => askYesNoOptional(lines, "Does this product send user data to third-party AI services as part of its own functionality?"));
+    // AI-assisted review (tier: byok) is an advanced opt-in, reached only via --enable-ai-review (or by
+    // hand-editing the config) -- the default flow never asks about it or mentions an API key. "managed"
+    // is likewise not offered here: it's still a real SrcLawyerTier value that config.ts and
+    // resolveProviderForTier.ts support, so a hand-edited tier: managed keeps its clear "not available
+    // yet" error at scan time, but nothing advertises a path to it.
+    const tier = flags.enableAiReview ? "byok" : "free";
     let apiKeyEnvVar = null;
     if (tier === "byok") {
-        apiKeyEnvVar = await ask(lines, "Which environment variable holds your Anthropic API key?", "ANTHROPIC_API_KEY");
+        apiKeyEnvVar = flags.apiKeyEnvVar ?? (await ask(lines, "Which environment variable holds your Anthropic API key?", DEFAULT_API_KEY_ENV_VAR));
+        if (!isValidEnvVarName(apiKeyEnvVar)) {
+            console.log(`\n"${apiKeyEnvVar}" is not a valid environment variable name; using ${DEFAULT_API_KEY_ENV_VAR} instead. Edit ${CONFIG_FILENAME} to change it.`);
+            apiKeyEnvVar = DEFAULT_API_KEY_ENV_VAR;
+        }
         console.log(`\nNote: only the name "${apiKeyEnvVar}" is written to ${CONFIG_FILENAME} — the key itself is never written to disk. It's read from your environment at scan time.`);
+        console.log(byokNetworkNotice(apiKeyEnvVar));
     }
-    rl.close();
     const config = {
         entityLocation: entityLocation || null,
         targetMarkets,
@@ -160,4 +175,14 @@ export async function runInit(root, flags = {}) {
     };
     writeConfig(root, config);
     console.log(`\nWrote ${CONFIG_FILENAME}. Run "srclawyer scan" to analyze this codebase.`);
+    let runScanNow = false;
+    if (isInteractive) {
+        const raw = await ask(lines, "Run a scan now and generate your policy?", "y");
+        runScanNow = /^y(es)?$/i.test(raw.trim());
+    }
+    rl.close();
+    if (runScanNow) {
+        console.log("");
+        await runScan(root, { json: false, out: "PRIVACY_POLICY.md", llm: true });
+    }
 }

package/dist/commands/scan.js

@@ -24,6 +24,12 @@ export async function runScan(root, options) {
     }
     const result = await scan(root);
     let findings = result.findings;
+    if (result.workspaceWarning)
+        console.error(result.workspaceWarning);
+    if (result.skippedFilesWarning)
+        console.error(result.skippedFilesWarning);
+    if (result.templateSyntaxNote)
+        console.error(result.templateSyntaxNote);
     if (result.maxZodFragmentBytes > 0) {
         console.error(`Largest Zod-schema fragment this scan: ${result.maxZodFragmentBytes} bytes (hard ceiling: ${MAX_FRAGMENT_BYTES} bytes).`);
     }
@@ -54,7 +60,7 @@ export async function runScan(root, options) {
                 if (config.tier === "byok" && config.llm) {
                     console.log(byokNetworkNotice(config.llm.apiKeyEnvVar));
                 }
-                console.log(`Sending ${ambiguous.length} ambiguous item(s) to ${provider.name}...`);
+                console.log(`Sending ${ambiguous.length} ambiguous item(s) to ${provider.name} (an opaque id, category, description and one redacted source line each)...`);
                 const resolved = await resolveAmbiguousFindings(ambiguous, provider, {
                     // Preserves today's exact override behavior bit-for-bit: a missing/legacy key already
                     // behaved as false before collectsChildrensData became nullable. Only the generated
@@ -71,14 +77,14 @@ export async function runScan(root, options) {
     const finalResult = { ...result, findings, ambiguousCount };
     const policyPath = resolve(root, options.out);
     writeFileSync(policyPath, generatePolicyMarkdown(finalResult, config, llmError, protectedLogicWarning));
-    if (result.unsupportedStackWarning || llmError || protectedLogicWarning || legacyConfigWarning) {
+    if (result.unsupportedStackWarning || result.workspaceWarning || result.skippedFilesWarning || llmError || protectedLogicWarning || legacyConfigWarning) {
         process.exitCode = 1;
     }
     if (options.json) {
         console.log(JSON.stringify({ ...finalResult, policyPath, llmError, protectedLogicWarning, legacyConfigWarning }, null, 2));
         return;
     }
-    console.log(formatText(finalResult));
+    console.log(formatText(finalResult, { aiReviewEnabled: config?.tier === "byok" }));
     if (protectedLogicWarning)
         console.log(`\n${protectedLogicWarning}`);
     if (legacyConfigWarning)

package/dist/engine/discoverFiles.d.ts

@@ -1,3 +1,4 @@
+import type { RootBoundary } from "./rootBoundary.js";
 export interface DiscoveredFiles {
     codeFiles: string[];
     pythonFiles: string[];
@@ -5,4 +6,4 @@ export interface DiscoveredFiles {
     schemaFiles: string[];
     manifestFiles: string[];
 }
-export declare function discoverFiles(root: string): Promise<DiscoveredFiles>;
+export declare function discoverFiles(root: string, boundary: RootBoundary): Promise<DiscoveredFiles>;

package/dist/engine/discoverFiles.js

@@ -1,4 +1,4 @@
-import fg from "fast-glob";
+import { globFiles } from "./globFiles.js";
 const CODE_PATTERNS = ["**/*.ts", "**/*.tsx", "**/*.js", "**/*.jsx", "**/*.mjs", "**/*.cjs"];
 const PYTHON_PATTERNS = ["**/*.py"];
 const HTML_PATTERNS = ["**/*.html", "**/*.htm"];
@@ -27,13 +27,13 @@ const PYTHON_IGNORE = [
     "**/test_*.py",
     "**/*_test.py",
 ];
-export async function discoverFiles(root) {
+export async function discoverFiles(root, boundary) {
     const [codeFiles, pythonFiles, htmlFiles, schemaFiles, manifestFiles] = await Promise.all([
-        fg(CODE_PATTERNS, { cwd: root, absolute: true, ignore: IGNORE, dot: false }),
-        fg(PYTHON_PATTERNS, { cwd: root, absolute: true, ignore: PYTHON_IGNORE, dot: false }),
-        fg(HTML_PATTERNS, { cwd: root, absolute: true, ignore: IGNORE, dot: false }),
-        fg(SCHEMA_PATTERNS, { cwd: root, absolute: true, ignore: ["**/node_modules/**"], dot: false }),
-        fg(MANIFEST_PATTERNS, { cwd: root, absolute: true, ignore: ["**/node_modules/**"], dot: false }),
+        globFiles(CODE_PATTERNS, { cwd: root, absolute: true, ignore: IGNORE, dot: false, boundary }),
+        globFiles(PYTHON_PATTERNS, { cwd: root, absolute: true, ignore: PYTHON_IGNORE, dot: false, boundary }),
+        globFiles(HTML_PATTERNS, { cwd: root, absolute: true, ignore: IGNORE, dot: false, boundary }),
+        globFiles(SCHEMA_PATTERNS, { cwd: root, absolute: true, ignore: ["**/node_modules/**"], dot: false, boundary }),
+        globFiles(MANIFEST_PATTERNS, { cwd: root, absolute: true, ignore: ["**/node_modules/**"], dot: false, boundary }),
     ]);
     return { codeFiles, pythonFiles, htmlFiles, schemaFiles, manifestFiles };
 }

package/dist/engine/discoverProjects.d.ts

@@ -1,3 +1,11 @@
+import type { RootBoundary } from "./rootBoundary.js";
 import type { ProjectManifest } from "./types.js";
-export declare function discoverProjects(root: string): Promise<ProjectManifest[]>;
+export interface ProjectDiscovery {
+    projects: ProjectManifest[];
+    /** Non-null when the scanned repo's own workspace config was rejected (malformed, hostile, or too
+     *  expensive to expand) and the default package locations were scanned instead -- surfaced on the
+     *  console and in the generated policy, since the scan may then be incomplete. */
+    workspaceWarning: string | null;
+}
+export declare function discoverProjects(root: string, boundary: RootBoundary): Promise<ProjectDiscovery>;
 export declare function describeProject(root: string, project: ProjectManifest): string;

package/dist/engine/discoverProjects.js

@@ -1,32 +1,85 @@
 import { readFileSync, existsSync } from "node:fs";
 import { dirname, join, relative } from "node:path";
-import fg from "fast-glob";
-export async function discoverProjects(root) {
+import { globFiles } from "./globFiles.js";
+import { validateWorkspacePatterns } from "./workspacePatterns.js";
+const DEFAULT_WORKSPACE_PATTERNS = ["packages/*", "apps/*"];
+function rejectedConfigWarning(reason) {
+    return `Workspace configuration in package.json was rejected (${reason}); the default package locations (${DEFAULT_WORKSPACE_PATTERNS.join(", ")}) were scanned instead, so this scan may be incomplete.`;
+}
+function readPackageJson(path, boundary) {
+    if (!boundary.allowsPath(path))
+        return { error: "package.json resolves outside the scanned root" };
+    try {
+        // A leading BOM (common from Windows editors) is valid in practice but not to JSON.parse.
+        const text = readFileSync(path, "utf8").replace(/^/, "");
+        return { value: JSON.parse(text) };
+    }
+    catch {
+        return { error: "package.json could not be parsed" };
+    }
+}
+// undefined means "no workspace config at all"; anything else (including a wrong type) is handed to
+// the validator, which rejects what isn't an array of well-formed patterns.
+function rawWorkspaces(pkg) {
+    const workspaces = typeof pkg === "object" && pkg !== null ? pkg.workspaces : undefined;
+    if (workspaces === undefined || workspaces === null)
+        return undefined;
+    if (typeof workspaces === "object" && !Array.isArray(workspaces))
+        return workspaces.packages ?? undefined;
+    return workspaces;
+}
+async function globManifests(patterns, root, boundary) {
+    return globFiles(patterns.map((p) => (p.startsWith("!") ? `!${join(p.slice(1), "package.json")}` : join(p, "package.json"))), { cwd: root, absolute: true, ignore: ["**/node_modules/**"], boundary });
+}
+export async function discoverProjects(root, boundary) {
     const rootPackageJsonPath = join(root, "package.json");
     if (!existsSync(rootPackageJsonPath)) {
-        return [{ root, kind: "unknown", manifestPath: null }];
-    }
-    const rootPackageJson = JSON.parse(readFileSync(rootPackageJsonPath, "utf8"));
-    const workspacePatterns = Array.isArray(rootPackageJson.workspaces)
-        ? rootPackageJson.workspaces
-        : rootPackageJson.workspaces?.packages;
-    const lernaPath = join(root, "lerna.json");
-    const nxPath = join(root, "nx.json");
-    const hasWorkspaceConfig = Boolean(workspacePatterns) || existsSync(lernaPath) || existsSync(nxPath);
+        return { projects: [{ root, kind: "unknown", manifestPath: null }], workspaceWarning: null };
+    }
+    const rootProject = { root, kind: "node", manifestPath: rootPackageJsonPath };
+    const parsed = readPackageJson(rootPackageJsonPath, boundary);
+    let workspaceWarning = null;
+    let patterns = null;
+    let hasWorkspaceConfig = existsSync(join(root, "lerna.json")) || existsSync(join(root, "nx.json"));
+    if ("error" in parsed) {
+        workspaceWarning = rejectedConfigWarning(parsed.error);
+        hasWorkspaceConfig = true;
+    }
+    else {
+        const raw = rawWorkspaces(parsed.value);
+        if (raw !== undefined) {
+            hasWorkspaceConfig = true;
+            const validation = validateWorkspacePatterns(raw);
+            if (validation.ok)
+                patterns = validation.patterns;
+            else
+                workspaceWarning = rejectedConfigWarning(validation.reason);
+        }
+    }
     if (!hasWorkspaceConfig) {
-        return [{ root, kind: "node", manifestPath: rootPackageJsonPath }];
+        return { projects: [rootProject], workspaceWarning: null };
+    }
+    let manifestPaths;
+    try {
+        manifestPaths = await globManifests(patterns ?? DEFAULT_WORKSPACE_PATTERNS, root, boundary);
+    }
+    catch {
+        // Belt and braces behind validateWorkspacePatterns: a pattern that still fails to expand must
+        // never take the whole scan down.
+        workspaceWarning = rejectedConfigWarning("the patterns could not be expanded");
+        try {
+            manifestPaths = await globManifests(DEFAULT_WORKSPACE_PATTERNS, root, boundary);
+        }
+        catch {
+            manifestPaths = [];
+        }
     }
-    const patterns = workspacePatterns ?? ["packages/*", "apps/*"];
-    const manifestPaths = await fg(patterns.map((p) => join(p, "package.json")), { cwd: root, absolute: true, ignore: ["**/node_modules/**"] });
     const projects = manifestPaths.map((manifestPath) => ({
         root: dirname(manifestPath),
         kind: "node",
         manifestPath,
     }));
-    if (projects.length === 0) {
-        return [{ root, kind: "node", manifestPath: rootPackageJsonPath }];
-    }
-    return projects;
+    return { projects: projects.length === 0 ? [rootProject] : projects, workspaceWarning };
 }
 export function describeProject(root, project) {
     const rel = relative(root, project.root);

package/dist/engine/globFiles.d.ts

@@ -0,0 +1,10 @@
+import type { RootBoundary } from "./rootBoundary.js";
+export interface GlobFilesOptions {
+    cwd: string;
+    absolute?: boolean;
+    ignore?: string[];
+    dot?: boolean;
+    /** When given, symlinks resolving outside the scanned root are neither walked nor returned. */
+    boundary?: RootBoundary;
+}
+export declare function globFiles(patterns: string | string[], options: GlobFilesOptions): Promise<string[]>;

package/dist/engine/globFiles.js

@@ -0,0 +1,12 @@
+import { glob } from "tinyglobby";
+// expandDirectories: false keeps tinyglobby's matching identical to the fast-glob behavior this
+// replaced (a pattern matches only what it names, never everything under a matched directory).
+// tinyglobby/picomatch compiles braces to regex alternation rather than expanding them, so it has no
+// equivalent of the `braces` package's stack-overflow/memory exposure (GHSA-vfj7-8cjw-p6xm); patterns
+// that come from a scanned repo are still bounded first, in workspacePatterns.ts.
+// Sorted so scan order (and therefore the order of findings in reports) doesn't depend on filesystem
+// traversal order, which neither fast-glob nor tinyglobby guarantees.
+export async function globFiles(patterns, options) {
+    const { boundary, ...globOptions } = options;
+    return (await glob(patterns, { expandDirectories: false, ...globOptions, ...(boundary ? { fs: boundary.globFs() } : {}) })).sort();
+}

package/dist/engine/languageDetection.d.ts

@@ -1 +1,2 @@
-export declare function detectUnsupportedStack(root: string): Promise<string | null>;
+import type { RootBoundary } from "./rootBoundary.js";
+export declare function detectUnsupportedStack(root: string, boundary: RootBoundary): Promise<string | null>;

package/dist/engine/languageDetection.js

@@ -1,6 +1,6 @@
 import { existsSync, readFileSync } from "node:fs";
 import { join } from "node:path";
-import fg from "fast-glob";
+import { globFiles } from "./globFiles.js";
 const IGNORE = ["**/node_modules/**", "**/dist/**", "**/build/**", "**/.git/**", "**/vendor/**", "**/venv/**", "**/.venv/**", "**/__pycache__/**"];
 // Python moved from UNSUPPORTED_LANGUAGES to here once Flask detection shipped (Stage A) -- it's a
 // supported *language* now, but only one framework within it is actually understood. A Python project
@@ -17,7 +17,7 @@ const FASTAPI_IMPORT = /\bfrom\s+fastapi\b|\bimport\s+fastapi\b/;
 function bannerize(message) {
     return `UNSUPPORTED CODEBASE — ${message} SrcLawyer only analyzes JavaScript/TypeScript, HTML, and Flask-based Python today; the findings below (if any) are incomplete and MUST NOT be treated as a full compliance scan.`;
 }
-async function detectFrameworkSignature(root) {
+async function detectFrameworkSignature(root, boundary) {
     if (existsSync(join(root, "manage.py"))) {
         return bannerize("Django detected (manage.py found at the project root) — Django support isn't built yet.");
     }
@@ -26,7 +26,7 @@ async function detectFrameworkSignature(root) {
     if (hasGemfile && hasRailsLayout) {
         return bannerize("Rails detected (Gemfile + a Rails app/config layout found).");
     }
-    const protoFiles = await fg("**/*.proto", { cwd: root, ignore: IGNORE, dot: false });
+    const protoFiles = await globFiles("**/*.proto", { cwd: root, ignore: IGNORE, dot: false, boundary });
     if (protoFiles.length > 0) {
         return bannerize(`gRPC/Protocol Buffers detected (${protoFiles.length} .proto file(s) found) — protobuf schemas are not parsed.`);
     }
@@ -53,10 +53,10 @@ async function detectUnsupportedPythonFramework(root, pythonFiles) {
     }
     return null;
 }
-async function detectDominantUnsupportedLanguage(root) {
+async function detectDominantUnsupportedLanguage(root, boundary) {
     const [supportedFiles, ...unsupportedGroups] = await Promise.all([
-        fg(SUPPORTED_EXTENSIONS.map((ext) => `**/*.${ext}`), { cwd: root, ignore: IGNORE, dot: false }),
-        ...UNSUPPORTED_LANGUAGES.map((lang) => fg(lang.extensions.map((ext) => `**/*.${ext}`), { cwd: root, ignore: IGNORE, dot: false })),
+        globFiles(SUPPORTED_EXTENSIONS.map((ext) => `**/*.${ext}`), { cwd: root, ignore: IGNORE, dot: false, boundary }),
+        ...UNSUPPORTED_LANGUAGES.map((lang) => globFiles(lang.extensions.map((ext) => `**/*.${ext}`), { cwd: root, ignore: IGNORE, dot: false, boundary })),
     ]);
     const counts = UNSUPPORTED_LANGUAGES.map((lang, i) => ({ label: lang.label, count: unsupportedGroups[i].length }));
     const dominant = counts.filter((c) => c.count > 0).sort((a, b) => b.count - a.count)[0];
@@ -64,15 +64,15 @@ async function detectDominantUnsupportedLanguage(root) {
         return null;
     return bannerize(`${dominant.label} detected as the primary language (${dominant.count} file(s), vs ${supportedFiles.length} supported file(s)).`);
 }
-export async function detectUnsupportedStack(root) {
-    const frameworkWarning = await detectFrameworkSignature(root);
+export async function detectUnsupportedStack(root, boundary) {
+    const frameworkWarning = await detectFrameworkSignature(root, boundary);
     if (frameworkWarning)
         return frameworkWarning;
-    const pythonFiles = await fg("**/*.py", { cwd: root, absolute: true, ignore: IGNORE, dot: false });
+    const pythonFiles = await globFiles("**/*.py", { cwd: root, absolute: true, ignore: IGNORE, dot: false, boundary });
     if (pythonFiles.length > 0) {
         const pythonFrameworkWarning = await detectUnsupportedPythonFramework(root, pythonFiles);
         if (pythonFrameworkWarning)
             return pythonFrameworkWarning;
     }
-    return detectDominantUnsupportedLanguage(root);
+    return detectDominantUnsupportedLanguage(root, boundary);
 }

package/dist/engine/rootBoundary.d.ts

@@ -0,0 +1,29 @@
+import { realpath, realpathSync } from "node:fs";
+/**
+ * The scanned root as a security boundary: a scanned repo is untrusted input, so no file whose REAL
+ * path resolves outside it may be read, and no directory symlink leaving it may be walked (a symlink
+ * to "/" or "~" would otherwise make a scan traverse the whole filesystem and report lines from files
+ * the repo's author chose). In-root symlinks, including loops that stay inside, keep working.
+ */
+export declare class RootBoundary {
+    readonly root: string;
+    readonly realRoot: string;
+    private readonly blocked;
+    private readonly resolved;
+    constructor(root: string);
+    isInsideRealPath(resolved: string): boolean;
+    /** Records a blocked link by its in-repo (lexical) path only -- never its target. */
+    private recordBlocked;
+    /** Read-time check, independent of how the path was discovered: the real path if it is inside the
+     *  root, else null (and an out-of-root link is recorded). Unresolvable paths fail closed. Cached. */
+    resolveInside(path: string): string | null;
+    allowsPath(path: string): boolean;
+    /** `fs` overrides for tinyglobby: a symlink resolving outside the root errors out, so it is neither
+     *  walked (directories) nor returned (files). */
+    globFs(): {
+        realpath: typeof realpath;
+        realpathSync: typeof realpathSync;
+    };
+    /** Relative in-repo paths of every blocked link, sorted; call after discovery and reading are done. */
+    blockedLinks(): string[];
+}