fix: stream foreground bash and close the small follow-up batch

sdk/py/harness_sdk.py

@@ -14,7 +14,7 @@
 from typing import Iterator, Optional
 
 MODELS = ["MiniMax-M2.5", "MiniMax-M2.7", "MiniMax-M3", "accounts/fireworks/models/deepseek-v4-flash", "accounts/fireworks/models/deepseek-v4-pro", "accounts/fireworks/models/glm-5p2", "accounts/fireworks/models/gpt-oss-120b", "accounts/fireworks/models/kimi-k2p6", "accounts/fireworks/models/kimi-k2p7-code", "accounts/fireworks/models/minimax-m3", "accounts/fireworks/models/qwen3p7-plus", "claude-fable-5", "claude-haiku-4-5", "claude-opus-4-5", "claude-opus-4-6", "claude-opus-4-7", "claude-opus-4-8", "claude-opus-4.8", "claude-opus-5", "claude-sonnet-4-5", "claude-sonnet-4-6", "claude-sonnet-4.6", "claude-sonnet-5", "deepseek-chat", "deepseek-reasoner", "deepseek-v4-flash", "deepseek-v4-pro", "fugu", "fugu-ultra", "fugu-ultra-20260615", "glm-4.5", "glm-4.7", "glm-5", "glm-5.2", "gpt-5-codex", "gpt-5.2", "gpt-5.3-codex-spark", "gpt-5.4", "gpt-5.4-mini", "gpt-5.4-pro", "gpt-5.5", "gpt-5.6", "gpt-5.6-luna", "gpt-5.6-sol", "gpt-5.6-terra", "grok-4.3", "grok-build", "k3", "kilo-auto/small", "kimi-for-coding", "kimi-for-coding-highspeed", "kimi-k2.6", "kimi-latest", "lmstudio", "mimo-v2-flash", "mimo-v2.5", "mimo-v2.5-pro", "mimo-v2.5-pro-ultraspeed", "minimax-m3", "mistral-medium-latest", "mlx-community/Qwen3.6-27B-OptiQ-4bit", "openai/gpt-oss-120b"]
-TOOLS = ["bash", "bash_output", "bash_kill", "read_file", "edit_file", "write_file", "webfetch", "skill", "codedb", "todo_write", "todo_read", "eval", "note_constraint", "ask_user", "attempt_completion", "load_tool_schemas", "clock_sleep", "subagent", "workflow", "agent_output", "learn_candidate", "peer_message", "imagegen"]
+TOOLS = ["bash", "bash_output", "bash_kill", "read_file", "edit_file", "write_file", "webfetch", "skill", "codedb", "todo_write", "todo_read", "eval", "note_constraint", "ask_user", "attempt_completion", "load_tool_schemas", "clock_sleep", "subagent", "workflow", "agent_output", "learn_candidate", "peer_message", "imagegen", "view_image"]
 PROVIDERS = ["anthropic", "codegraff", "deepseek", "openai", "minimax", "xiaomi", "kilo", "groq", "mistral", "kimi", "moonshot", "xai", "zai", "fugu", "fireworks", "mlx", "lmstudio", "codex"]
 
 

sdk/ts/harness.ts

@@ -12,7 +12,7 @@ import type { Readable, Writable } from "node:stream";
 export const HARNESS_VERSION = "0.12";
 
 export type ModelName = "MiniMax-M2.5" | "MiniMax-M2.7" | "MiniMax-M3" | "accounts/fireworks/models/deepseek-v4-flash" | "accounts/fireworks/models/deepseek-v4-pro" | "accounts/fireworks/models/glm-5p2" | "accounts/fireworks/models/gpt-oss-120b" | "accounts/fireworks/models/kimi-k2p6" | "accounts/fireworks/models/kimi-k2p7-code" | "accounts/fireworks/models/minimax-m3" | "accounts/fireworks/models/qwen3p7-plus" | "claude-fable-5" | "claude-haiku-4-5" | "claude-opus-4-5" | "claude-opus-4-6" | "claude-opus-4-7" | "claude-opus-4-8" | "claude-opus-4.8" | "claude-opus-5" | "claude-sonnet-4-5" | "claude-sonnet-4-6" | "claude-sonnet-4.6" | "claude-sonnet-5" | "deepseek-chat" | "deepseek-reasoner" | "deepseek-v4-flash" | "deepseek-v4-pro" | "fugu" | "fugu-ultra" | "fugu-ultra-20260615" | "glm-4.5" | "glm-4.7" | "glm-5" | "glm-5.2" | "gpt-5-codex" | "gpt-5.2" | "gpt-5.3-codex-spark" | "gpt-5.4" | "gpt-5.4-mini" | "gpt-5.4-pro" | "gpt-5.5" | "gpt-5.6" | "gpt-5.6-luna" | "gpt-5.6-sol" | "gpt-5.6-terra" | "grok-4.3" | "grok-build" | "k3" | "kilo-auto/small" | "kimi-for-coding" | "kimi-for-coding-highspeed" | "kimi-k2.6" | "kimi-latest" | "lmstudio" | "mimo-v2-flash" | "mimo-v2.5" | "mimo-v2.5-pro" | "mimo-v2.5-pro-ultraspeed" | "minimax-m3" | "mistral-medium-latest" | "mlx-community/Qwen3.6-27B-OptiQ-4bit" | "openai/gpt-oss-120b" | (string & {});
-export type ToolName = "bash" | "bash_output" | "bash_kill" | "read_file" | "edit_file" | "write_file" | "webfetch" | "skill" | "codedb" | "todo_write" | "todo_read" | "eval" | "note_constraint" | "ask_user" | "attempt_completion" | "load_tool_schemas" | "clock_sleep" | "subagent" | "workflow" | "agent_output" | "learn_candidate" | "peer_message" | "imagegen";
+export type ToolName = "bash" | "bash_output" | "bash_kill" | "read_file" | "edit_file" | "write_file" | "webfetch" | "skill" | "codedb" | "todo_write" | "todo_read" | "eval" | "note_constraint" | "ask_user" | "attempt_completion" | "load_tool_schemas" | "clock_sleep" | "subagent" | "workflow" | "agent_output" | "learn_candidate" | "peer_message" | "imagegen" | "view_image";
 export type ProviderId = "anthropic" | "codegraff" | "deepseek" | "openai" | "minimax" | "xiaomi" | "kilo" | "groq" | "mistral" | "kimi" | "moonshot" | "xai" | "zai" | "fugu" | "fireworks" | "mlx" | "lmstudio" | "codex";
 
 /** Events streamed by `harness --json` (one JSON object per stdout line), each
@@ -661,4 +661,4 @@ export async function* runAgent(opts: RunAgentOptions): AsyncGenerator<Event> {
 }
 
 export const MODELS: ModelName[] = ["MiniMax-M2.5", "MiniMax-M2.7", "MiniMax-M3", "accounts/fireworks/models/deepseek-v4-flash", "accounts/fireworks/models/deepseek-v4-pro", "accounts/fireworks/models/glm-5p2", "accounts/fireworks/models/gpt-oss-120b", "accounts/fireworks/models/kimi-k2p6", "accounts/fireworks/models/kimi-k2p7-code", "accounts/fireworks/models/minimax-m3", "accounts/fireworks/models/qwen3p7-plus", "claude-fable-5", "claude-haiku-4-5", "claude-opus-4-5", "claude-opus-4-6", "claude-opus-4-7", "claude-opus-4-8", "claude-opus-4.8", "claude-opus-5", "claude-sonnet-4-5", "claude-sonnet-4-6", "claude-sonnet-4.6", "claude-sonnet-5", "deepseek-chat", "deepseek-reasoner", "deepseek-v4-flash", "deepseek-v4-pro", "fugu", "fugu-ultra", "fugu-ultra-20260615", "glm-4.5", "glm-4.7", "glm-5", "glm-5.2", "gpt-5-codex", "gpt-5.2", "gpt-5.3-codex-spark", "gpt-5.4", "gpt-5.4-mini", "gpt-5.4-pro", "gpt-5.5", "gpt-5.6", "gpt-5.6-luna", "gpt-5.6-sol", "gpt-5.6-terra", "grok-4.3", "grok-build", "k3", "kilo-auto/small", "kimi-for-coding", "kimi-for-coding-highspeed", "kimi-k2.6", "kimi-latest", "lmstudio", "mimo-v2-flash", "mimo-v2.5", "mimo-v2.5-pro", "mimo-v2.5-pro-ultraspeed", "minimax-m3", "mistral-medium-latest", "mlx-community/Qwen3.6-27B-OptiQ-4bit", "openai/gpt-oss-120b"];
-export const TOOLS: ToolName[] = ["bash", "bash_output", "bash_kill", "read_file", "edit_file", "write_file", "webfetch", "skill", "codedb", "todo_write", "todo_read", "eval", "note_constraint", "ask_user", "attempt_completion", "load_tool_schemas", "clock_sleep", "subagent", "workflow", "agent_output", "learn_candidate", "peer_message", "imagegen"];
+export const TOOLS: ToolName[] = ["bash", "bash_output", "bash_kill", "read_file", "edit_file", "write_file", "webfetch", "skill", "codedb", "todo_write", "todo_read", "eval", "note_constraint", "ask_user", "attempt_completion", "load_tool_schemas", "clock_sleep", "subagent", "workflow", "agent_output", "learn_candidate", "peer_message", "imagegen", "view_image"];

sdk/ts/remote.ts

@@ -10,7 +10,7 @@
 export const HARNESS_VERSION = "0.12";
 
 export type ModelName = "MiniMax-M2.5" | "MiniMax-M2.7" | "MiniMax-M3" | "accounts/fireworks/models/deepseek-v4-flash" | "accounts/fireworks/models/deepseek-v4-pro" | "accounts/fireworks/models/glm-5p2" | "accounts/fireworks/models/gpt-oss-120b" | "accounts/fireworks/models/kimi-k2p6" | "accounts/fireworks/models/kimi-k2p7-code" | "accounts/fireworks/models/minimax-m3" | "accounts/fireworks/models/qwen3p7-plus" | "claude-fable-5" | "claude-haiku-4-5" | "claude-opus-4-5" | "claude-opus-4-6" | "claude-opus-4-7" | "claude-opus-4-8" | "claude-opus-4.8" | "claude-opus-5" | "claude-sonnet-4-5" | "claude-sonnet-4-6" | "claude-sonnet-4.6" | "claude-sonnet-5" | "deepseek-chat" | "deepseek-reasoner" | "deepseek-v4-flash" | "deepseek-v4-pro" | "fugu" | "fugu-ultra" | "fugu-ultra-20260615" | "glm-4.5" | "glm-4.7" | "glm-5" | "glm-5.2" | "gpt-5-codex" | "gpt-5.2" | "gpt-5.3-codex-spark" | "gpt-5.4" | "gpt-5.4-mini" | "gpt-5.4-pro" | "gpt-5.5" | "gpt-5.6" | "gpt-5.6-luna" | "gpt-5.6-sol" | "gpt-5.6-terra" | "grok-4.3" | "grok-build" | "k3" | "kilo-auto/small" | "kimi-for-coding" | "kimi-for-coding-highspeed" | "kimi-k2.6" | "kimi-latest" | "lmstudio" | "mimo-v2-flash" | "mimo-v2.5" | "mimo-v2.5-pro" | "mimo-v2.5-pro-ultraspeed" | "minimax-m3" | "mistral-medium-latest" | "mlx-community/Qwen3.6-27B-OptiQ-4bit" | "openai/gpt-oss-120b" | (string & {});
-export type ToolName = "bash" | "bash_output" | "bash_kill" | "read_file" | "edit_file" | "write_file" | "webfetch" | "skill" | "codedb" | "todo_write" | "todo_read" | "eval" | "note_constraint" | "ask_user" | "attempt_completion" | "load_tool_schemas" | "clock_sleep" | "subagent" | "workflow" | "agent_output" | "learn_candidate" | "peer_message" | "imagegen";
+export type ToolName = "bash" | "bash_output" | "bash_kill" | "read_file" | "edit_file" | "write_file" | "webfetch" | "skill" | "codedb" | "todo_write" | "todo_read" | "eval" | "note_constraint" | "ask_user" | "attempt_completion" | "load_tool_schemas" | "clock_sleep" | "subagent" | "workflow" | "agent_output" | "learn_candidate" | "peer_message" | "imagegen" | "view_image";
 export type ProviderId = "anthropic" | "codegraff" | "deepseek" | "openai" | "minimax" | "xiaomi" | "kilo" | "groq" | "mistral" | "kimi" | "moonshot" | "xai" | "zai" | "fugu" | "fireworks" | "mlx" | "lmstudio" | "codex";
 
 /** Events streamed by the bridge (same `--json` contract as the stdio SDK).
@@ -508,4 +508,4 @@ export async function* runAgentRemote(opts: RunAgentRemoteOptions): AsyncGenerat
 }
 
 export const MODELS: ModelName[] = ["MiniMax-M2.5", "MiniMax-M2.7", "MiniMax-M3", "accounts/fireworks/models/deepseek-v4-flash", "accounts/fireworks/models/deepseek-v4-pro", "accounts/fireworks/models/glm-5p2", "accounts/fireworks/models/gpt-oss-120b", "accounts/fireworks/models/kimi-k2p6", "accounts/fireworks/models/kimi-k2p7-code", "accounts/fireworks/models/minimax-m3", "accounts/fireworks/models/qwen3p7-plus", "claude-fable-5", "claude-haiku-4-5", "claude-opus-4-5", "claude-opus-4-6", "claude-opus-4-7", "claude-opus-4-8", "claude-opus-4.8", "claude-opus-5", "claude-sonnet-4-5", "claude-sonnet-4-6", "claude-sonnet-4.6", "claude-sonnet-5", "deepseek-chat", "deepseek-reasoner", "deepseek-v4-flash", "deepseek-v4-pro", "fugu", "fugu-ultra", "fugu-ultra-20260615", "glm-4.5", "glm-4.7", "glm-5", "glm-5.2", "gpt-5-codex", "gpt-5.2", "gpt-5.3-codex-spark", "gpt-5.4", "gpt-5.4-mini", "gpt-5.4-pro", "gpt-5.5", "gpt-5.6", "gpt-5.6-luna", "gpt-5.6-sol", "gpt-5.6-terra", "grok-4.3", "grok-build", "k3", "kilo-auto/small", "kimi-for-coding", "kimi-for-coding-highspeed", "kimi-k2.6", "kimi-latest", "lmstudio", "mimo-v2-flash", "mimo-v2.5", "mimo-v2.5-pro", "mimo-v2.5-pro-ultraspeed", "minimax-m3", "mistral-medium-latest", "mlx-community/Qwen3.6-27B-OptiQ-4bit", "openai/gpt-oss-120b"];
-export const TOOLS: ToolName[] = ["bash", "bash_output", "bash_kill", "read_file", "edit_file", "write_file", "webfetch", "skill", "codedb", "todo_write", "todo_read", "eval", "note_constraint", "ask_user", "attempt_completion", "load_tool_schemas", "clock_sleep", "subagent", "workflow", "agent_output", "learn_candidate", "peer_message", "imagegen"];
+export const TOOLS: ToolName[] = ["bash", "bash_output", "bash_kill", "read_file", "edit_file", "write_file", "webfetch", "skill", "codedb", "todo_write", "todo_read", "eval", "note_constraint", "ask_user", "attempt_completion", "load_tool_schemas", "clock_sleep", "subagent", "workflow", "agent_output", "learn_candidate", "peer_message", "imagegen", "view_image"];

src/agent_request_policy.zig

@@ -92,6 +92,9 @@ pub fn refreshLoginKeyBeforeSend(self: *Agent) void {
     // Only adopt on a real CHANGE: an unchanged credential must not churn the
     // session arena, nor hand the WS transport latch back on every request.
     if (std.mem.eql(u8, fresh.key, self.provider.api_key)) return;
+    // #404: a failed persist left the live token only in memory. Do not
+    // replace it with an older on-disk credential.
+    if (oauth.persistFailed() and !std.mem.eql(u8, fresh.key, self.provider.api_key)) return;
     adoptFreshAuth(self, fresh);
 }
 

src/agent_tools.zig

@@ -135,6 +135,7 @@ pub fn runTools(self: *Agent, calls: []const ToolCall) ![]ExecResult {
             .subagent_cross_provider = self.subagent_cross_provider,
             .registry = if (self.sub) null else self.registry,
             .from_sub = self.sub,
+            .out = self.out,
             .has_eval = self.eval_cmd != null,
             .approvals = self.approvals,
             .tracer = self.tracer,

src/credential_store.zig

@@ -31,7 +31,9 @@ pub const private_dir: Io.File.Permissions = if (Io.File.Permissions.has_executa
 /// `private_file` for anything holding a secret; pass `.default_file` to keep
 /// the ordinary umask-governed behaviour of a plain `createFile`.
 pub fn replaceFile(io: Io, dir: Io.Dir, sub_path: []const u8, bytes: []const u8, permissions: Io.File.Permissions) !void {
-    var atomic = try dir.createFileAtomic(io, sub_path, .{ .permissions = permissions, .replace = true });
+    var path_buf: [std.fs.max_path_bytes]u8 = undefined;
+    const dest = if (dir.readLink(io, sub_path, &path_buf)) |n| path_buf[0..n] else |_| sub_path;
+    var atomic = try dir.createFileAtomic(io, dest, .{ .permissions = permissions, .replace = true });
     defer atomic.deinit(io);
     if (Io.File.Permissions.has_executable_bit) {
         if (permissions != .default_file) {
@@ -196,3 +198,22 @@ test "writeOAuth: the credential file is 0600 inside 0700 directories" {
     try std.testing.expectEqualStrings("refresh-1", parsed.object.get("refresh_token").?.string);
     try std.testing.expectEqual(@as(i64, 1234), parsed.object.get("expires_at").?.integer);
 }
+
+test "#405: replaceFile writes through a symlink instead of replacing it" {
+    if (@import("builtin").os.tag == .windows) return;
+    const io = std.testing.io;
+    var tmp = std.testing.tmpDir(.{});
+    defer tmp.cleanup();
+    var real_buf: [std.fs.max_path_bytes]u8 = undefined;
+    const base = real_buf[0..try tmp.dir.realPath(io, &real_buf)];
+    const target = try std.fmt.allocPrint(std.testing.allocator, "{s}/real.json", .{base});
+    defer std.testing.allocator.free(target);
+    const link = try std.fmt.allocPrint(std.testing.allocator, "{s}/settings.json", .{base});
+    defer std.testing.allocator.free(link);
+    try Io.Dir.cwd().writeFile(io, .{ .sub_path = target, .data = "old\n" });
+    try Io.Dir.cwd().symLink(io, target, link, .{});
+    try replaceFile(io, Io.Dir.cwd(), link, "new\n", .default_file);
+    const through = try Io.Dir.cwd().readFileAlloc(io, target, std.testing.allocator, .limited(64));
+    defer std.testing.allocator.free(through);
+    try std.testing.expectEqualStrings("new\n", through);
+}

src/exec.zig

@@ -54,6 +54,7 @@ const noSymlinkEscape = approvals_mod.noSymlinkEscape;
 const jobs = @import("jobs.zig");
 const runCapped = jobs.runCapped;
 const runCappedWithOptions = jobs.runCappedWithOptions;
+const exec_bash_stream = @import("exec_bash_stream.zig");
 const toolRunOptions = jobs.toolRunOptions; // #266/#198: own the child's process group
 const spawnJob = jobs.spawnJob;
 const jobOutput = jobs.jobOutput;
@@ -76,7 +77,8 @@ const no_local_tools = @import("no_local_tools.zig"); // #330: the hard --no-loc
 const native_fold = @import("native_fold.zig"); // folded native power tools: layer-2 refusal until load_tool_schemas unfolds
 const vision = @import("vision.zig"); // read_file stages images like MCP image results (#249)
 const input_util = @import("input_util.zig");
-const imagegen = @import("imagegen.zig"); // #352: the codex-gated image tool (advertising lives in schema.zig/tool_gates.zig)
+const imagegen = @import("imagegen.zig");
+const view_image = @import("view_image.zig");
 
 /// Wall-clock ceiling for one *subagent* bash command. Subagents run on pool
 /// threads with no TTY, so there is no Esc to kill a runaway command — without
@@ -325,13 +327,11 @@ fn execToolInner(ctx: ToolCtx, call: ToolCall) !ToolOutput {
         }
         const sh = shellArgv(cmd);
         const deadline: u64 = if (ctx.from_sub) subagent_bash_deadline_ms else 0;
-        // #276 P0-1: a worktree-isolated agent's bash calls run pinned to its
-        // own worktree — via std.process.Child.Cwd, per spawn, never a
-        // process-wide chdir — so parallel siblings never share a cwd.
-        // #266/#198: toolRunOptions also gives the command its own process
-        // group, so an Esc cancel or the deadline kills what it spawned (ssh,
-        // xcodebuild) instead of leaving it running against a dead turn.
-        const run = try runCappedWithOptions(gpa, io, &sh, bash_stdout_cap, bash_stderr_cap, deadline, toolRunOptions(ctx.agent_cwd));
+        // Isolated cwd + process group.
+        var opts = toolRunOptions(ctx.agent_cwd);
+        var stream = exec_bash_stream.Ctx{ .io = io, .w = ctx.out };
+        exec_bash_stream.attach(&opts, !ctx.from_sub, &stream);
+        const run = try runCappedWithOptions(gpa, io, &sh, bash_stdout_cap, bash_stderr_cap, deadline, opts);
         defer gpa.free(run.stdout);
         defer gpa.free(run.stderr);
 
@@ -545,9 +545,8 @@ fn execToolInner(ctx: ToolCtx, call: ToolCall) !ToolOutput {
     // Loads one SKILL.md body (or lists them). Rescans on every call, so a
     // skill written this session is loadable without a restart.
     if (std.mem.eql(u8, call.name, "skill")) return skill_docs.execSkill(gpa, io, input);
-    // #352: codex-gated. execImagegen answers a call that was never advertised
-    // (an unavailable session) with the same honest error it gives the model.
     if (std.mem.eql(u8, call.name, imagegen.tool_name)) return imagegen.execImagegen(ctx, input);
+    if (std.mem.eql(u8, call.name, view_image.name)) return view_image.exec(ctx, input);
     if (std.mem.eql(u8, call.name, "subagent")) return execSubagent(ctx, input);
     if (std.mem.eql(u8, call.name, "workflow")) return execWorkflow(ctx, input);
     if (std.mem.eql(u8, call.name, "agent_output")) {
@@ -594,6 +593,7 @@ test "internal learning respects the parent privacy ceiling" {
 }
 
 test { // main.zig is at the 600-line cap; exec.zig is these modules' importer, so the compiled-in references live here (the reach check diffs the test binary, not which file holds the line)
+    _ = exec_bash_stream;
     _ = @import("codedbpro_report.zig");
     _ = @import("tool_balance.zig");
 }

src/exec_bash_stream.zig

@@ -0,0 +1,60 @@
+//! Live foreground-bash output (#472). runCapped already polls pipes every
+//! 200ms; this file is the UI half — write each new chunk so a long build
+//! does not look like a hang.
+
+const std = @import("std");
+const Io = std.Io;
+
+const main_mod = @import("main.zig");
+const jobs = @import("jobs.zig");
+
+pub const Ctx = struct {
+    io: Io,
+    w: ?*Io.Writer,
+};
+
+pub fn attach(opts: *jobs.CappedRunOptions, live: bool, stream: *Ctx) void {
+    if (!live) return;
+    opts.stream = &emit;
+    opts.stream_ctx = stream;
+}
+
+pub fn emit(ctx: ?*anyopaque, which: u8, chunk: []const u8) void {
+    if (chunk.len == 0) return;
+    const c: *const Ctx = @ptrCast(@alignCast(ctx orelse return));
+    if (main_mod.json_mode) {
+        const w = main_mod.g_out orelse return;
+        main_mod.g_gui_mu.lockUncancelable(c.io);
+        defer main_mod.g_gui_mu.unlock(c.io);
+        w.writeAll("{\"type\":\"tool_output\",\"name\":\"bash\",\"stream\":\"") catch return;
+        w.writeAll(if (which == 0) "stdout" else "stderr") catch return;
+        w.writeAll("\",\"text\":") catch return;
+        writeJsonString(w, chunk) catch return;
+        w.writeAll("}\n") catch return;
+        w.flush() catch {};
+        return;
+    }
+    const w = c.w orelse main_mod.g_out orelse return;
+    w.writeAll(chunk) catch return;
+    w.flush() catch {};
+}
+
+fn writeJsonString(w: *Io.Writer, s: []const u8) !void {
+    try w.writeByte('"');
+    for (s) |c| switch (c) {
+        '"' => try w.writeAll("\\\""),
+        '\\' => try w.writeAll("\\\\"),
+        '\n' => try w.writeAll("\\n"),
+        '\r' => try w.writeAll("\\r"),
+        '\t' => try w.writeAll("\\t"),
+        else => if (c < 0x20) try w.print("\\u{d:0>4}", .{c}) else try w.writeByte(c),
+    };
+    try w.writeByte('"');
+}
+
+test "#472: writeJsonString escapes a newline" {
+    var buf: [64]u8 = undefined;
+    var w: Io.Writer = .fixed(&buf);
+    try writeJsonString(&w, "a\nb");
+    try std.testing.expectEqualStrings("\"a\\nb\"", w.buffered());
+}