jb/paseo
1
0
Fork 0
Code Issues 1 Pull Requests Actions Packages Projects Releases Wiki Activity

refactor: simplify tool call display and path handling

Browse Source 
- Remove displayName from tool calls, use name directly with icon
- Add medium font weight for tool call badges
- Strip workspace root from file paths at parsing level using Zod schema
- Add accentBorder to tool call badges and git diff file blocks
- Fix user message alignment and vertical spacing
- Deduplicate stripCwdPrefix into tool-call-parsers.ts

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
Mohamed Boudra
2026-01-07 12:38:04 +07:00
co-authored by Claude Opus 4.5
parent 4f3b2317ab
commit 66881511b6
18 changed files with 475 additions and 445 deletions
Download Patch File Download Diff File Expand all files Collapse all files
packages/server/src/server/agent/activity-curator.ts
+29 -11
View File
@@ -1,7 +1,19 @@
import type { AgentTimelineItem } from "./agent-sdk-types.js";
import type { AgentTimelineItem, ToolCallKind } from "./agent-sdk-types.js";
const DEFAULT_MAX_ITEMS = 40;
/**
* Derive tool kind from the tool name for rendering purposes.
*/
function getToolKind(name: string): ToolCallKind {
const lower = name.toLowerCase();
if (lower === "read" || lower === "read_file") return "read";
if (lower === "edit" || lower === "write" || lower === "apply_patch") return "edit";
if (lower === "bash" || lower === "shell") return "execute";
if (lower === "grep" || lower === "glob" || lower === "web_search") return "search";
return "other";
}
function appendText(buffer: string, text: string): string {
const normalized = text.trim();
if (!normalized) {
@@ -50,6 +62,13 @@ function extractWebQuery(value: unknown): string {
return value.query;
}
function extractCommand(value: unknown): string {
if (!isObject(value) || typeof value.command !== "string") {
return "";
}
return value.command;
}
/**
* Convert normalized agent timeline items into a concise text summary.
*/
@@ -84,13 +103,12 @@ export function curateAgentActivity(
break;
case "tool_call": {
flushBuffers(lines, buffers);
const label =
item.displayName ??
(item.server && item.tool ? `${item.server}.${item.tool}` : item.tool ?? item.server ?? "Tool");
const status = item.status ? ` ${item.status}` : "";
if (item.kind === "execute" || item.server === "command") {
lines.push(`[Command: ${label}]${status}`);
} else if (item.kind === "edit" || item.server === "file_change") {
const kind = getToolKind(item.name);
if (kind === "execute") {
const command = extractCommand(item.input);
lines.push(`[Command: ${command || item.name}]${status}`);
} else if (kind === "edit") {
const files = extractFileChanges(item.output);
if (files.length > 0) {
lines.push("[File Changes]");
@@ -98,13 +116,13 @@ export function curateAgentActivity(
lines.push(`- (${file.kind}) ${file.path}`);
}
} else {
lines.push(`[Edit] ${label}${status}`);
lines.push(`[Edit] ${item.name}${status}`);
}
} else if (item.kind === "search" || item.server === "web_search") {
} else if (kind === "search") {
const query = extractWebQuery(item.input);
lines.push(`[Web Search] ${query || label}`);
lines.push(`[Web Search] ${query || item.name}`);
} else {
lines.push(`[Tool ${item.server}.${item.tool}]${status}`);
lines.push(`[Tool ${item.name}]${status}`);
}
break;
}
packages/server/src/server/agent/agent-sdk-types.ts
+24 -12
View File
@@ -55,22 +55,34 @@ export type AgentUsage = {
totalCostUsd?: number;
};
/**
* Tool call kind categories for UI rendering hints.
* Derived from the tool name, not sent over the wire.
*/
export type ToolCallKind = "read" | "edit" | "execute" | "search" | "other";
/**
* Clean tool call structure.
* - `name`: Tool identifier (e.g., "Read", "Bash", "Edit", "shell", "read_file", "apply_patch")
* - `input`: Tool input parameters
* - `output`: Tool result
* - `error`: Error if tool failed
*/
export interface ToolCallTimelineItem {
type: "tool_call";
name: string;
callId?: string;
status?: string;
input?: unknown;
output?: unknown;
error?: unknown;
}
export type AgentTimelineItem =
| { type: "user_message"; text: string; messageId?: string }
| { type: "assistant_message"; text: string }
| { type: "reasoning"; text: string }
| {
type: "tool_call";
server: string;
tool: string;
status?: string;
callId?: string;
displayName?: string;
kind?: string;
input?: unknown;
output?: unknown;
error?: unknown;
}
| ToolCallTimelineItem
| { type: "todo"; items: { text: string; completed: boolean }[] }
| { type: "error"; message: string };
packages/server/src/server/agent/providers/claude-agent.test.ts
+16 -24
View File
@@ -88,21 +88,16 @@ function extractCommandText(input: unknown): string | null {
}
function isSleepCommandToolCall(item: ToolCallItem): boolean {
const display = typeof item.displayName === "string" ? item.displayName.toLowerCase() : "";
if (display.includes("sleep 60")) {
return true;
}
const inputCommand = extractCommandText(item.input)?.toLowerCase() ?? "";
return inputCommand.includes("sleep 60");
}
function isPermissionCommandToolCall(item: ToolCallItem): boolean {
if (item.server === "permission") {
if (item.name === "permission_request") {
return false;
}
const display = typeof item.displayName === "string" ? item.displayName.toLowerCase() : "";
const inputCommand = extractCommandText(item.input)?.toLowerCase() ?? "";
return display.includes("permission.txt") || inputCommand.includes("permission.txt");
return inputCommand.includes("permission.txt");
}
type AgentMcpServerHandle = {
@@ -354,7 +349,7 @@ describe("ClaudeAgentClient (SDK integration)", () => {
const config = buildConfig(cwd, { maxThinkingTokens: 2048 });
const session = await client.createSession(config);
let pendingDisplay: string | null = null;
let pendingCommand: string | null = null;
const events = session.stream("Run the exact command `pwd` via Bash and stop.");
try {
@@ -363,10 +358,10 @@ describe("ClaudeAgentClient (SDK integration)", () => {
if (
event.type === "timeline" &&
event.item.type === "tool_call" &&
event.item.server.toLowerCase().includes("bash") &&
event.item.name.toLowerCase().includes("bash") &&
event.item.status === "pending"
) {
pendingDisplay = event.item.displayName ?? null;
pendingCommand = extractCommandText(event.item.input);
}
if (event.type === "turn_completed" || event.type === "turn_failed") {
break;
@@ -377,8 +372,8 @@ describe("ClaudeAgentClient (SDK integration)", () => {
rmSync(cwd, { recursive: true, force: true });
}
expect(pendingDisplay).toBeTruthy();
expect(pendingDisplay?.toLowerCase()).toContain("pwd");
expect(pendingCommand).toBeTruthy();
expect(pendingCommand?.toLowerCase()).toContain("pwd");
},
150_000
);
@@ -418,9 +413,8 @@ describe("ClaudeAgentClient (SDK integration)", () => {
);
const commandEvents = toolCalls.filter(
(item) =>
(item.kind === "execute" || item.server === "command") &&
typeof item.displayName === "string" &&
!item.displayName.startsWith("permission:")
item.name.toLowerCase().includes("bash") &&
item.name !== "permission_request"
);
const fileChangeEvent = toolCalls.find((item) => {
// Check for file changes in structured output.files array
@@ -444,7 +438,7 @@ describe("ClaudeAgentClient (SDK integration)", () => {
});
const sawPwdCommand = commandEvents.some(
(item) => (item.displayName ?? "").toLowerCase().includes("pwd") && item.status === "completed"
(item) => (extractCommandText(item.input) ?? "").toLowerCase().includes("pwd") && item.status === "completed"
);
expect(completed).toBe(true);
@@ -906,7 +900,8 @@ describe("ClaudeAgentClient (SDK integration)", () => {
const liveState = hydrateStreamState(liveTimelineUpdates);
const liveSnapshots = extractAgentToolSnapshots(liveState);
const commandTool = liveSnapshots.find((snapshot) =>
(snapshot.data.displayName ?? "").toLowerCase().includes("pwd")
snapshot.data.name.toLowerCase().includes("bash") &&
(extractCommandText(snapshot.data.input) ?? "").toLowerCase().includes("pwd")
);
const editTool = liveSnapshots.find((snapshot) =>
rawContainsText(snapshot.data.result, "hydrate-proof.txt")
@@ -953,8 +948,8 @@ describe("ClaudeAgentClient (SDK integration)", () => {
({ live, hydrated }) => {
expect(rawContainsText(live.result, cwd)).toBe(true);
expect(rawContainsText(hydrated.result, cwd)).toBe(true);
expect((live.displayName ?? "").toLowerCase()).toContain("pwd");
expect((hydrated.displayName ?? "").toLowerCase()).toContain("pwd");
expect((extractCommandText(live.input) ?? "").toLowerCase()).toContain("pwd");
expect((extractCommandText(hydrated.input) ?? "").toLowerCase()).toContain("pwd");
}
);
assertHydratedReplica(
@@ -1165,8 +1160,7 @@ function buildToolSnapshotKey(data: AgentToolCallData, fallbackId: string): stri
if (normalized) {
return normalized;
}
const display = typeof data.displayName === "string" && data.displayName.trim().length > 0 ? data.displayName.trim() : fallbackId;
return `${data.server}:${data.tool}:${display}`;
return `${data.provider}:${data.name}:${fallbackId}`;
}
function assertHydratedReplica(
@@ -1179,9 +1173,7 @@ function assertHydratedReplica(
const hydrated = hydratedMap.get(liveSnapshot.key);
expect(hydrated).toBeTruthy();
expect(hydrated?.status).toBe(liveSnapshot.data.status);
expect(hydrated?.server).toBe(liveSnapshot.data.server);
expect(hydrated?.tool).toBe(liveSnapshot.data.tool);
expect(hydrated?.displayName).toBe(liveSnapshot.data.displayName);
expect(hydrated?.name).toBe(liveSnapshot.data.name);
expect(predicate(hydrated!)).toBe(true);
if (hydrated && extraAssertions) {
extraAssertions({ live: liveSnapshot.data, hydrated });
packages/server/src/server/agent/providers/claude-agent.ts
+9 -53
View File
@@ -619,12 +619,9 @@ class ClaudeAgentSession implements AgentSession {
if (pending.request.kind === "plan") {
await this.setMode("acceptEdits");
this.pushToolCall({
server: "plan",
tool: "plan_approval",
name: "plan_approval",
status: "granted",
callId: pending.request.id,
displayName: "Plan approved",
kind: "plan",
});
}
const result: PermissionResult = {
@@ -634,12 +631,9 @@ class ClaudeAgentSession implements AgentSession {
};
pending.resolve(result);
this.pushToolCall({
server: "permission",
tool: pending.request.name,
name: pending.request.name,
status: "granted",
callId: pending.request.id,
displayName: pending.request.title ?? pending.request.name,
kind: "permission",
input: pending.request.input,
});
} else {
@@ -650,12 +644,9 @@ class ClaudeAgentSession implements AgentSession {
};
pending.resolve(result);
this.pushToolCall({
server: "permission",
tool: pending.request.name,
name: pending.request.name,
status: "denied",
callId: pending.request.id,
displayName: pending.request.title ?? pending.request.name,
kind: "permission",
input: pending.request.input,
});
}
@@ -1039,12 +1030,9 @@ class ClaudeAgentSession implements AgentSession {
};
this.pushToolCall({
server: "permission",
tool: toolName,
name: toolName,
status: "requested",
callId: requestId,
displayName: request.title ?? toolName,
kind: "permission",
input,
});
@@ -1067,12 +1055,9 @@ class ClaudeAgentSession implements AgentSession {
cleanup();
const error = new Error("Permission request timed out");
this.pushToolCall({
server: "permission",
tool: toolName,
name: toolName,
status: "denied",
callId: requestId,
displayName: request.title ?? toolName,
kind: "permission",
input,
});
this.pushEvent({
@@ -1138,25 +1123,6 @@ class ClaudeAgentSession implements AgentSession {
this.enqueueTimeline(item);
}
private getToolKind(classification?: ToolUseClassification): string | undefined {
switch (classification) {
case "command":
return "execute";
case "file_change":
return "edit";
case "generic":
default:
return "tool";
}
}
private buildToolDisplayName(entry?: ToolUseCacheEntry): string | undefined {
if (!entry) {
return undefined;
}
return entry.commandText ?? entry.name;
}
private pushEvent(event: AgentStreamEvent) {
if (this.eventQueue) {
this.eventQueue.push(event);
@@ -1315,12 +1281,9 @@ class ClaudeAgentSession implements AgentSession {
this.toolUseCache.set(entry.id, entry);
this.pushToolCall(
{
server: entry.server,
tool: entry.name,
name: entry.name,
status: "pending",
callId: entry.id,
displayName: this.buildToolDisplayName(entry),
kind: this.getToolKind(entry.classification),
input: entry.input ?? this.normalizeToolInput(block.input),
},
items
@@ -1329,8 +1292,7 @@ class ClaudeAgentSession implements AgentSession {
private handleToolResult(block: ClaudeContentChunk, items: AgentTimelineItem[]): void {
const entry = typeof block.tool_use_id === "string" ? this.toolUseCache.get(block.tool_use_id) : undefined;
const server = entry?.server ?? block.server ?? "tool";
const tool = entry?.name ?? block.tool_name ?? "tool";
const toolName = entry?.name ?? block.tool_name ?? "tool";
const status = block.is_error ? "failed" : "completed";
// Extract output from block.content (SDK always returns content in string form)
@@ -1338,12 +1300,9 @@ class ClaudeAgentSession implements AgentSession {
this.pushToolCall(
{
server,
tool,
name: toolName,
status,
callId: typeof block.tool_use_id === "string" ? block.tool_use_id : undefined,
displayName: this.buildToolDisplayName(entry),
kind: this.getToolKind(entry?.classification),
input: entry?.input,
output,
error: block.is_error ? block : undefined,
@@ -1577,12 +1536,9 @@ class ClaudeAgentSession implements AgentSession {
this.applyToolInput(entry, normalized);
this.toolUseCache.set(toolId, entry);
this.pushToolCall({
server: entry.server,
tool: entry.name,
name: entry.name,
status: "pending",
callId: toolId,
displayName: this.buildToolDisplayName(entry),
kind: this.getToolKind(entry.classification),
input: normalized,
});
}
packages/server/src/server/agent/providers/codex-mcp-agent.test.ts
-4
View File
@@ -250,10 +250,6 @@ function stringifyUnknown(value: unknown): string {
}
function isSleepCommandToolCall(item: ToolCallItem): boolean {
const display = typeof item.displayName === "string" ? item.displayName.toLowerCase() : "";
if (display.includes("sleep 60")) {
return true;
}
const inputText = commandTextFromInput(item.input);
if (!inputText) {
return false;
packages/server/src/server/agent/providers/codex-mcp-agent.ts
+15 -70
View File
@@ -1954,13 +1954,6 @@ function shouldReportCommandError(input: {
return false;
}
function buildFileChangeSummary(files: { path: string; kind: string }[]): string {
if (files.length === 1) {
return `${files[0].kind}: ${files[0].path}`;
}
return `${files.length} file changes`;
}
function extractContentText(content: unknown): string | null {
if (!Array.isArray(content)) {
return null;
@@ -3105,12 +3098,9 @@ class CodexMcpAgentSession implements AgentSession {
type: "timeline",
provider: CODEX_PROVIDER,
item: createToolCallTimelineItem({
server: "permission",
tool: pending.request.name,
name: pending.request.name,
status,
callId: pending.request.id,
displayName: pending.request.title ? pending.request.title : pending.request.name,
kind: "permission",
input: pending.request.input,
}),
});
@@ -3415,12 +3405,9 @@ class CodexMcpAgentSession implements AgentSession {
type: "timeline",
provider: CODEX_PROVIDER,
item: createToolCallTimelineItem({
server: "permission",
tool: request.name,
name: request.name,
status: "requested",
callId: request.id,
displayName: request.title ? request.title : request.name,
kind: "permission",
input: request.input,
}),
});
@@ -3535,12 +3522,9 @@ class CodexMcpAgentSession implements AgentSession {
type: "timeline",
provider: CODEX_PROVIDER,
item: createToolCallTimelineItem({
server: "file_change",
tool: "apply_patch",
name: "apply_patch",
status: success ? "completed" : "failed",
callId,
displayName: buildFileChangeSummary(summaryFiles),
kind: "edit",
input: { files: summaryFiles },
output: { files: pendingChanges, message: parsedOutputText, success },
}),
@@ -3607,19 +3591,15 @@ class CodexMcpAgentSession implements AgentSession {
if (!callId) {
throw new Error("exec_command_begin missing call_id");
}
const commandText = normalizeCommand(parsedEvent.command);
const fileRead = extractFileReadFromParsedCmd(parsedEvent.parsedCmd);
if (fileRead) {
this.emitEvent({
type: "timeline",
provider: CODEX_PROVIDER,
item: createToolCallTimelineItem({
server: "file",
tool: "read_file",
name: "read_file",
status: "running",
callId,
displayName: `Read: ${fileRead.name}`,
kind: "read",
input: { path: fileRead.path },
}),
});
@@ -3628,12 +3608,9 @@ class CodexMcpAgentSession implements AgentSession {
type: "timeline",
provider: CODEX_PROVIDER,
item: createToolCallTimelineItem({
server: "command",
tool: "shell",
name: "shell",
status: "running",
callId,
displayName: commandText,
kind: "execute",
input: { command: parsedEvent.command, cwd: parsedEvent.cwd },
}),
});
@@ -3709,12 +3686,9 @@ class CodexMcpAgentSession implements AgentSession {
type: "timeline",
provider: CODEX_PROVIDER,
item: createToolCallTimelineItem({
server: "file",
tool: "read_file",
name: "read_file",
status: failed ? "failed" : "completed",
callId,
displayName: `Read: ${fileRead.name}`,
kind: "read",
input: { path: fileRead.path },
output: {
type: "read_file",
@@ -3749,12 +3723,9 @@ class CodexMcpAgentSession implements AgentSession {
type: "timeline",
provider: CODEX_PROVIDER,
item: createToolCallTimelineItem({
server: "command",
tool: "shell",
name: "shell",
status: failed ? "failed" : "completed",
callId,
displayName: commandText,
kind: "execute",
input: { command: parsedEvent.command, cwd: parsedEvent.cwd },
output: structuredOutput,
}),
@@ -3791,12 +3762,9 @@ class CodexMcpAgentSession implements AgentSession {
type: "timeline",
provider: CODEX_PROVIDER,
item: createToolCallTimelineItem({
server: "file_change",
tool: "apply_patch",
name: "apply_patch",
status: "running",
callId,
displayName: buildFileChangeSummary(files),
kind: "edit",
input: { changes: parsedEvent.changes, files },
}),
});
@@ -3852,12 +3820,9 @@ class CodexMcpAgentSession implements AgentSession {
type: "timeline",
provider: CODEX_PROVIDER,
item: createToolCallTimelineItem({
server: "file_change",
tool: "apply_patch",
name: "apply_patch",
status: parsedEvent.success ? "completed" : "failed",
callId,
displayName: buildFileChangeSummary(summaryFiles),
kind: "edit",
input: { files: summaryFiles },
output,
}),
@@ -3881,12 +3846,9 @@ class CodexMcpAgentSession implements AgentSession {
type: "timeline",
provider: CODEX_PROVIDER,
item: createToolCallTimelineItem({
server: parsedEvent.server,
tool: parsedEvent.tool,
name: `${parsedEvent.server}.${parsedEvent.tool}`,
status: success ? "completed" : "failed",
callId: parsedEvent.callId,
displayName: `${parsedEvent.server}.${parsedEvent.tool}`,
kind: "tool",
input,
output: normalizedOutput,
}),
@@ -4059,12 +4021,9 @@ class CodexMcpAgentSession implements AgentSession {
commandOutput.exitCode = resolvedExitCode;
}
return createToolCallTimelineItem({
server: "command",
tool: "shell",
name: "shell",
status: item.status,
callId: item.callId,
displayName: command,
kind: "execute",
input: { command: item.command, cwd: item.cwd },
output: commandOutput,
error: item.error,
@@ -4089,12 +4048,9 @@ class CodexMcpAgentSession implements AgentSession {
? "running"
: "completed";
return createToolCallTimelineItem({
server: "file_change",
tool: "apply_patch",
name: "apply_patch",
status,
callId: item.callId,
displayName: buildFileChangeSummary(summaryFiles),
kind: "edit",
input: { files: summaryFiles },
output: { files: changes },
});
@@ -4104,16 +4060,12 @@ class CodexMcpAgentSession implements AgentSession {
return null;
}
const readItem = item as ReadFileThreadItem;
const displayName = `Read ${readItem.path}`;
const output =
readItem.content !== undefined ? readItem.content : readItem.output;
return createToolCallTimelineItem({
server: "file_read",
tool: "read_file",
name: "read_file",
status: readItem.status,
callId: readItem.callId,
displayName,
kind: "read",
input: readItem.input ? readItem.input : { path: readItem.path },
output,
});
@@ -4123,12 +4075,9 @@ class CodexMcpAgentSession implements AgentSession {
return null;
}
return createToolCallTimelineItem({
server: item.server,
tool: item.tool,
name: `${item.server}.${item.tool}`,
status: item.status,
callId: item.callId,
displayName: `${item.server}.${item.tool}`,
kind: "tool",
input: item.input,
output: item.output,
});
@@ -4137,16 +4086,12 @@ class CodexMcpAgentSession implements AgentSession {
if (eventType && eventType !== "item.completed") {
return null;
}
const displayName = `Web search: ${item.query}`;
const output =
item.results !== undefined ? item.results : item.output;
return createToolCallTimelineItem({
server: "web_search",
tool: "web_search",
name: "web_search",
status: item.status,
callId: item.callId,
displayName,
kind: "search",
input: item.input ? item.input : { query: item.query },
output,
});
packages/server/src/server/daemon.e2e.test.ts
+230 -102
View File
@@ -8,7 +8,7 @@ import {
useTempClaudeConfigDir,
} from "./test-utils/index.js";
import type { AgentTimelineItem } from "./agent/agent-sdk-types.js";
import type { AgentSnapshotPayload } from "./messages.js";
import type { AgentSnapshotPayload, SessionOutboundMessage } from "./messages.js";
function tmpCwd(): string {
return mkdtempSync(path.join(tmpdir(), "daemon-e2e-"));
@@ -3117,82 +3117,81 @@ describe("daemon E2E", () => {
});
describe("tool call structure", () => {
// Helper to extract and dedupe tool calls by callId, keeping the last (most complete) version
function extractToolCalls(
queue: SessionOutboundMessage[],
agentId: string
): AgentTimelineItem[] {
const byCallId = new Map<string, AgentTimelineItem>();
const noCallId: AgentTimelineItem[] = [];
for (const m of queue) {
if (
m.type === "agent_stream" &&
m.payload.agentId === agentId &&
m.payload.event.type === "timeline" &&
m.payload.event.item.type === "tool_call"
) {
const tc = m.payload.event.item;
if (tc.callId) {
byCallId.set(tc.callId, tc);
} else {
noCallId.push(tc);
}
}
}
return [...byCallId.values(), ...noCallId];
}
// Helper to log tool call structure in a consistent format
function logToolCall(prefix: string, tc: AgentTimelineItem): void {
if (tc.type !== "tool_call") return;
console.log(
`[${prefix}]`,
JSON.stringify({
name: tc.name,
callId: tc.callId,
status: tc.status,
hasInput: tc.input !== undefined,
hasOutput: tc.output !== undefined,
})
);
}
test(
"Claude agent tool calls have expected structure",
"Claude agent: Read tool",
async () => {
const cwd = tmpCwd();
// Create Claude agent with bypass permissions
const agent = await ctx.client.createAgent({
provider: "claude",
cwd,
title: "Tool Structure Test - Claude",
title: "Claude Read Test",
modeId: "bypassPermissions",
});
expect(agent.provider).toBe("claude");
ctx.client.clearMessageQueue();
// Prompt that triggers a Read tool call
await ctx.client.sendMessage(
agent.id,
"Read the file /etc/hosts and tell me how many lines it has. Be brief."
);
const finalState = await ctx.client.waitForAgentIdle(agent.id, 120000);
expect(finalState.status).toBe("idle");
// Extract tool_call timeline items
const queue = ctx.client.getMessageQueue();
const toolCalls: AgentTimelineItem[] = [];
for (const m of queue) {
if (
m.type === "agent_stream" &&
m.payload.agentId === agent.id &&
m.payload.event.type === "timeline" &&
m.payload.event.item.type === "tool_call"
) {
toolCalls.push(m.payload.event.item);
}
}
await ctx.client.waitForAgentIdle(agent.id, 120000);
const toolCalls = extractToolCalls(ctx.client.getMessageQueue(), agent.id);
expect(toolCalls.length).toBeGreaterThan(0);
// Log and verify structure for each tool call
for (const tc of toolCalls) {
expect(tc.type).toBe("tool_call");
// Current structure has: server, tool, displayName, kind
expect(typeof tc.server).toBe("string");
expect(typeof tc.tool).toBe("string");
console.log(
"[CLAUDE TOOL_CALL]",
JSON.stringify({
server: tc.server,
tool: tc.tool,
displayName: tc.displayName,
kind: tc.kind,
callId: tc.callId,
status: tc.status,
hasInput: tc.input !== undefined,
hasOutput: tc.output !== undefined,
})
);
logToolCall("CLAUDE_READ", tc);
}
// Find a Read tool call specifically
const readCall = toolCalls.find(
(tc) =>
tc.server === "Read" ||
tc.tool === "Read" ||
tc.displayName === "Read"
);
const readCall = toolCalls.find((tc) => tc.type === "tool_call" && tc.name === "Read");
expect(readCall).toBeDefined();
expect(readCall?.name).toBe("Read");
expect(readCall?.input).toBeDefined();
// Cleanup
await ctx.client.deleteAgent(agent.id);
rmSync(cwd, { recursive: true, force: true });
},
@@ -3200,80 +3199,209 @@ describe("daemon E2E", () => {
);
test(
"Codex agent tool calls have expected structure",
"Claude agent: Bash tool",
async () => {
const cwd = tmpCwd();
// Create Codex agent with full access
const agent = await ctx.client.createAgent({
provider: "codex",
provider: "claude",
cwd,
title: "Tool Structure Test - Codex",
modeId: "full-access",
title: "Claude Bash Test",
modeId: "bypassPermissions",
});
expect(agent.provider).toBe("codex");
ctx.client.clearMessageQueue();
// Prompt that triggers a shell command
await ctx.client.sendMessage(
agent.id,
"Run `echo hello` and tell me what it outputs. Be brief."
);
const finalState = await ctx.client.waitForAgentIdle(agent.id, 120000);
expect(finalState.status).toBe("idle");
// Extract tool_call timeline items
const queue = ctx.client.getMessageQueue();
const toolCalls: AgentTimelineItem[] = [];
for (const m of queue) {
if (
m.type === "agent_stream" &&
m.payload.agentId === agent.id &&
m.payload.event.type === "timeline" &&
m.payload.event.item.type === "tool_call"
) {
toolCalls.push(m.payload.event.item);
}
}
await ctx.client.waitForAgentIdle(agent.id, 120000);
const toolCalls = extractToolCalls(ctx.client.getMessageQueue(), agent.id);
expect(toolCalls.length).toBeGreaterThan(0);
// Log and verify structure for each tool call
for (const tc of toolCalls) {
expect(tc.type).toBe("tool_call");
// Current structure has: server, tool, displayName, kind
expect(typeof tc.server).toBe("string");
expect(typeof tc.tool).toBe("string");
console.log(
"[CODEX TOOL_CALL]",
JSON.stringify({
server: tc.server,
tool: tc.tool,
displayName: tc.displayName,
kind: tc.kind,
callId: tc.callId,
status: tc.status,
hasInput: tc.input !== undefined,
hasOutput: tc.output !== undefined,
})
);
logToolCall("CLAUDE_BASH", tc);
}
// Find a shell/execute tool call
const shellCall = toolCalls.find(
(tc) =>
tc.kind === "execute" ||
tc.server?.includes("shell") ||
tc.displayName?.includes("echo")
const bashCall = toolCalls.find((tc) => tc.type === "tool_call" && tc.name === "Bash");
expect(bashCall).toBeDefined();
expect(bashCall?.name).toBe("Bash");
expect(bashCall?.input).toBeDefined();
// Command text should be in input.command
const bashInput = bashCall?.input as { command?: string } | undefined;
expect(bashInput?.command).toContain("echo");
await ctx.client.deleteAgent(agent.id);
rmSync(cwd, { recursive: true, force: true });
},
180000
);
test(
"Claude agent: Edit tool",
async () => {
const cwd = tmpCwd();
const testFile = path.join(cwd, "test.txt");
writeFileSync(testFile, "hello world\n");
const agent = await ctx.client.createAgent({
provider: "claude",
cwd,
title: "Claude Edit Test",
modeId: "bypassPermissions",
});
ctx.client.clearMessageQueue();
await ctx.client.sendMessage(
agent.id,
`Edit the file ${testFile} and change "hello" to "goodbye". Be brief.`
);
await ctx.client.waitForAgentIdle(agent.id, 120000);
const toolCalls = extractToolCalls(ctx.client.getMessageQueue(), agent.id);
expect(toolCalls.length).toBeGreaterThan(0);
for (const tc of toolCalls) {
logToolCall("CLAUDE_EDIT", tc);
}
const editCall = toolCalls.find((tc) => tc.type === "tool_call" && tc.name === "Edit");
expect(editCall).toBeDefined();
expect(editCall?.input).toBeDefined();
await ctx.client.deleteAgent(agent.id);
rmSync(cwd, { recursive: true, force: true });
},
180000
);
test(
"Codex agent: shell command",
async () => {
const cwd = tmpCwd();
const agent = await ctx.client.createAgent({
provider: "codex",
cwd,
title: "Codex Shell Test",
modeId: "full-access",
});
ctx.client.clearMessageQueue();
await ctx.client.sendMessage(
agent.id,
"Run `echo hello` and tell me what it outputs. Be brief."
);
await ctx.client.waitForAgentIdle(agent.id, 120000);
const toolCalls = extractToolCalls(ctx.client.getMessageQueue(), agent.id);
expect(toolCalls.length).toBeGreaterThan(0);
for (const tc of toolCalls) {
logToolCall("CODEX_SHELL", tc);
}
const shellCall = toolCalls.find((tc) => tc.type === "tool_call" && tc.name === "shell");
expect(shellCall).toBeDefined();
expect(shellCall?.name).toBe("shell");
expect(shellCall?.input).toBeDefined();
// Command text should be in input.command
const shellInput = shellCall?.input as { command?: string } | undefined;
expect(shellInput?.command).toContain("echo");
await ctx.client.deleteAgent(agent.id);
rmSync(cwd, { recursive: true, force: true });
},
180000
);
test(
"Codex agent: file read",
async () => {
const cwd = tmpCwd();
const agent = await ctx.client.createAgent({
provider: "codex",
cwd,
title: "Codex Read Test",
modeId: "full-access",
});
ctx.client.clearMessageQueue();
await ctx.client.sendMessage(
agent.id,
"Read the file /etc/hosts and tell me how many lines it has. Be brief."
);
await ctx.client.waitForAgentIdle(agent.id, 120000);
const toolCalls = extractToolCalls(ctx.client.getMessageQueue(), agent.id);
expect(toolCalls.length).toBeGreaterThan(0);
for (const tc of toolCalls) {
logToolCall("CODEX_READ", tc);
}
// Codex may use shell cat or file read
const readCall = toolCalls.find(
(tc) => tc.type === "tool_call" && tc.name === "read_file"
);
if (readCall) {
expect(readCall.name).toBe("read_file");
}
await ctx.client.deleteAgent(agent.id);
rmSync(cwd, { recursive: true, force: true });
},
180000
);
test(
"Codex agent: file edit",
async () => {
const cwd = tmpCwd();
const testFile = path.join(cwd, "test.txt");
writeFileSync(testFile, "hello world\n");
const agent = await ctx.client.createAgent({
provider: "codex",
cwd,
title: "Codex Edit Test",
modeId: "full-access",
});
ctx.client.clearMessageQueue();
await ctx.client.sendMessage(
agent.id,
`Edit the file ${testFile} and change "hello" to "goodbye". Be brief.`
);
await ctx.client.waitForAgentIdle(agent.id, 120000);
const toolCalls = extractToolCalls(ctx.client.getMessageQueue(), agent.id);
expect(toolCalls.length).toBeGreaterThan(0);
for (const tc of toolCalls) {
logToolCall("CODEX_EDIT", tc);
}
// Codex uses apply_patch for edits
const editCall = toolCalls.find(
(tc) => tc.type === "tool_call" && tc.name === "apply_patch"
);
if (editCall) {
expect(editCall.name).toBe("apply_patch");
}
// Cleanup
await ctx.client.deleteAgent(agent.id);
rmSync(cwd, { recursive: true, force: true });
},
packages/server/src/server/messages.ts
+2 -5
View File
@@ -142,12 +142,9 @@ export const AgentTimelineItemPayloadSchema: z.ZodType<AgentTimelineItem> =
}),
z.object({
type: z.literal("tool_call"),
server: z.string(),
tool: z.string(),
status: z.string().optional(),
name: z.string(),
callId: z.string().optional(),
displayName: z.string().optional(),
kind: z.string().optional(),
status: z.string().optional(),
input: z.unknown().optional(),
output: z.unknown().optional(),
error: z.unknown().optional(),