feat: safe Evidence tree enumeration and bounded blob streaming (P6)

backend/src/workspaces/git-repository.ts

@@ -16,6 +16,12 @@ export interface GitStatus {
   lastError?: WorkspaceErrorCode;
 }
 
+export interface EvidenceTreeObject {
+  mode: "100644" | "100755";
+  oid: string;
+  posixPath: string;
+}
+
 export class WorkspaceRegistryError extends Error {
   constructor(readonly code: WorkspaceErrorCode, message: string) {
     super(message);
@@ -266,16 +272,69 @@ export class GitWorkspaceRepository {
       throw new WorkspaceRegistryError("workspace_invalid", "Workspace annotations object is invalid");
     }
     const blobId = match[3];
-    const contents = await this.gitBlobBuffer(blobId, 16 * 1024 * 1024);
+    const contents = await this.gitBlobBytes(blobId, 16 * 1024 * 1024, "annotations");
     if (!isValidUtf8(contents)) {
       throw new WorkspaceRegistryError("workspace_invalid", "Workspace annotations object is not valid UTF-8");
     }
     return { blobId, contents };
   }
 
-  private async gitBlobBuffer(objectId: string, maxBytes: number): Promise<Buffer> {
+  /**
+   * Recursively enumerate a canonical `<id>/evidence` tree at an exact commit as regular Git blobs.
+   * Symlinks (120000), gitlinks (160000), non-regular modes, non-blob types, traversal/absolute/
+   * duplicate/cross-namespace paths, and NUL/newline-bearing names are refused.
+   */
+  async evidenceTreeObjects(revision: string, id: string): Promise<EvidenceTreeObject[]> {
+    if (!/^[0-9a-f]{40}$/.test(revision)) {
+      throw new WorkspaceRegistryError("workspace_invalid", "Workspace Evidence revision is invalid");
+    }
+    if (!/^[a-z][a-z0-9-]{2,62}$/.test(id)) {
+      throw new WorkspaceRegistryError("workspace_invalid", "Workspace Evidence path is invalid");
+    }
+    const prefix = `${id}/evidence`;
+    const listing = await this.git(["ls-tree", "-r", "-z", "--full-tree", revision, "--", prefix]);
+    const entries = listing.split("\0").filter((entry) => entry.length > 0);
+    const seen = new Set<string>();
+    const objects: EvidenceTreeObject[] = [];
+    for (const entry of entries) {
+      const tab = entry.lastIndexOf("\t");
+      if (tab < 0) throw new WorkspaceRegistryError("workspace_invalid", "Workspace Evidence object is invalid");
+      const name = entry.slice(tab + 1);
+      const meta = entry.slice(0, tab);
+      const match = /^([0-9]{6}) (blob|commit|tree) ([0-9a-f]{40})$/.exec(meta);
+      if (match === null || match[2] !== "blob" || (match[1] !== "100644" && match[1] !== "100755")) {
+        throw new WorkspaceRegistryError("workspace_invalid", "Workspace Evidence object is invalid");
+      }
+      if (name === prefix) {
+        // The Evidence root resolves to a single regular blob (or symlink/gitlink already refused above).
+        throw new WorkspaceRegistryError("workspace_invalid", "Workspace Evidence root is invalid");
+      }
+      if (!name.startsWith(`${prefix}/`)) {
+        throw new WorkspaceRegistryError("workspace_invalid", "Workspace Evidence object escapes its namespace");
+      }
+      const rel = name.slice(prefix.length + 1);
+      if (rel.length === 0 || rel.includes("\0") || rel.includes("\n") || rel.includes("\r")) {
+        throw new WorkspaceRegistryError("workspace_invalid", "Workspace Evidence path is invalid");
+      }
+      const segments = rel.split("/");
+      if (segments.some((segment) => segment === "" || segment === "." || segment === "..")) {
+        throw new WorkspaceRegistryError("workspace_invalid", "Workspace Evidence path is invalid");
+      }
+      if (seen.has(rel)) throw new WorkspaceRegistryError("workspace_invalid", "Workspace Evidence path is duplicated");
+      seen.add(rel);
+      objects.push({ mode: match[1] as "100644" | "100755", oid: match[3], posixPath: rel });
+    }
+    return objects;
+  }
+
+  /** Read one Evidence blob with a per-object byte bound. */
+  evidenceBlobBytes(objectId: string, maxBytes: number): Promise<Buffer> {
+    return this.gitBlobBytes(objectId, maxBytes, "Evidence");
+  }
+
+  private async gitBlobBytes(objectId: string, maxBytes: number, label: string): Promise<Buffer> {
     if (!/^[0-9a-f]{40}$/.test(objectId)) {
-      throw new WorkspaceRegistryError("workspace_invalid", "Workspace annotations object is invalid");
+      throw new WorkspaceRegistryError("workspace_invalid", `Workspace ${label} object is invalid`);
     }
     try {
       const { stdout } = await execFileAsync(
@@ -289,14 +348,14 @@ export class GitWorkspaceRepository {
         },
       );
       if (stdout.length > maxBytes) {
-        throw new WorkspaceRegistryError("workspace_invalid", "Workspace annotations object is too large");
+        throw new WorkspaceRegistryError("workspace_invalid", `Workspace ${label} object is too large`);
       }
       return stdout;
     } catch (error) {
       if (error instanceof WorkspaceRegistryError) throw error;
       const detail = error instanceof Error ? error.message : "";
       if (/maxBuffer|stdout maxBuffer/i.test(detail)) {
-        throw new WorkspaceRegistryError("workspace_invalid", "Workspace annotations object is too large");
+        throw new WorkspaceRegistryError("workspace_invalid", `Workspace ${label} object is too large`);
       }
       throw this.sanitizeGitError(error);
     }