feat(brain): Live-Tool-Events im Gedanken-Stream

Proxy-Patch hookt Claude-CLI `assistant`-Events: bei jedem tool_use-
Block (Bash, Read, Edit, Grep, ...) wird per HTTP-POST an die Bridge
gemeldet. Bridge spiegelt das als `agent_activity tool=<name>` an die
RVS-Clients. App- und Diagnostic-Gedanken-Stream zeigen damit live mit
was ARIA gerade macht — vorher kam pro Brain-Call nur EIN „💭 denkt"
am Anfang und EIN „✓ fertig" am Ende.

Drei neue Bausteine:
- proxy-patches/routes.js: kompletter Replacement der npm-Version mit
  `_attachToolHook(subprocess)` — feuert pro tool_use-Block ein HTTP-
  POST an http://aria-bridge:8090/internal/agent-activity (URL via
  ARIA_TOOL_HOOK_URL Env-Variable ueberschreibbar). Fire-and-forget,
  fail-open — Brain-Call bricht NICHT ab wenn Bridge mal nicht da ist.
- docker-compose.yml: vierter cp-Schritt im proxy-Service kopiert
  routes.js ueber die npm-Version (analog zu openai-to-cli + cli-to-
  openai).
- bridge/aria_bridge.py: neuer `/internal/agent-activity`-Endpoint im
  bestehenden _serve_internal_http. Plus _emit_activity hat jetzt
  force=True-Param damit wiederholte gleiche Tool-Aufrufe (3x Bash in
  Folge) als drei Eintraege im Stream sichtbar bleiben.

App + Diagnostic: pushThought-Dedup laesst tool-Events durch (3x Bash
hintereinander gibt 3 Eintraege im Gedanken-Stream).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
2026-05-15 11:07:39 +02:00
parent 8ca899aaf5
commit bf3dc635d9
5 changed files with 345 additions and 7 deletions
+5 -2
View File
@@ -1024,9 +1024,12 @@ const ChatScreen: React.FC = () => {
const tool = (message.payload.tool as string) || '';
setAgentActivity({ activity, tool });
// In den Gedanken-Stream einfuegen. Dedup gegen identische Folge-
// Events (z.B. zwei mal 'thinking' direkt hintereinander).
// Events (z.B. zwei mal 'thinking' direkt hintereinander). Tool-
// Events NIE dedupen — wenn ARIA dreimal Bash hintereinander ruft,
// sollen alle drei sichtbar sein.
const key = `${activity}|${tool}`;
if (key !== lastThoughtKeyRef.current) {
const isTool = activity === 'tool';
if (isTool || key !== lastThoughtKeyRef.current) {
lastThoughtKeyRef.current = key;
setThoughts(prev => {
const next = [...prev, { ts: Date.now(), activity, tool }];
+26 -3
View File
@@ -2519,17 +2519,22 @@ class ARIABridge:
status = await asyncio.get_event_loop().run_in_executor(None, _do_request)
logger.info("[cancel] Diagnostic /api/cancel: %s", status)
async def _emit_activity(self, activity: str, tool: str = "") -> None:
async def _emit_activity(self, activity: str, tool: str = "", force: bool = False) -> None:
"""Sendet agent_activity an die App — nur wenn sich der State geaendert hat.
Trailing Agent-Events nach chat:final werden 3s lang unterdrueckt
(nur 'idle' kommt immer durch)."""
(nur 'idle' kommt immer durch).
force=True: kein State-Dedup — wird vom Proxy-Tool-Hook genutzt
damit auch wiederholte gleiche Tool-Aufrufe (z.B. 3x Bash
hintereinander) im Gedanken-Stream als eigene Eintraege sichtbar
bleiben."""
if activity != "idle" and self._last_chat_final_at > 0:
since_final = asyncio.get_event_loop().time() - self._last_chat_final_at
if since_final < 3.0:
return
state = (activity, tool)
if state == self._last_activity_state:
if not force and state == self._last_activity_state:
return
self._last_activity_state = state
await self._send_to_rvs({
@@ -2677,6 +2682,24 @@ class ARIABridge:
self._handle_trigger_fired(reply, trigger_name, ttype, events)
)
await _send_response(writer, 200, {"ok": True})
elif method == "POST" and path == "/internal/agent-activity":
# Vom Proxy gefeuert bei jedem Claude-Code-tool_use-Event
# (Bash, Read, Edit, Grep, ...). Wir spiegeln das als
# RVS agent_activity an App+Diagnostic damit der Gedanken-
# Stream live mitlaufen kann.
try:
data = json.loads(body.decode("utf-8", "ignore"))
except Exception as exc:
await _send_response(writer, 400, {"error": f"bad json: {exc}"})
return
tool = (data.get("tool") or "").strip()
if not tool:
await _send_response(writer, 400, {"error": "tool erforderlich"})
return
# Force-emit (kein Dedup): User soll JEDEN Tool-Call sehen
# selbst wenn derselbe Name zweimal in Folge kommt.
asyncio.create_task(self._emit_activity("tool", tool, force=True))
await _send_response(writer, 200, {"ok": True})
elif method == "POST" and path == "/internal/delete-chat-message":
try:
data = json.loads(body.decode("utf-8", "ignore"))
+4 -2
View File
@@ -2252,9 +2252,11 @@
}
function pushThought(activity, tool) {
// Dedup gegen direkt aufeinanderfolgende identische Events
// Dedup gegen direkt aufeinanderfolgende identische Events. Tool-
// Events NIE dedupen — drei Bash-Calls in Folge sollen drei Eintraege
// ergeben, nicht einen.
const key = `${activity}|${tool || ''}`;
if (key === lastThoughtKey) return;
if (activity !== 'tool' && key === lastThoughtKey) return;
lastThoughtKey = key;
thoughtStream.push({ ts: Date.now(), activity, tool: tool || '' });
if (thoughtStream.length > MAX_THOUGHTS) thoughtStream = thoughtStream.slice(-MAX_THOUGHTS);
+1
View File
@@ -15,6 +15,7 @@ services:
sed -i 's/const DEFAULT_TIMEOUT = 300000;/const DEFAULT_TIMEOUT = 1200000;/' $$DIST/subprocess/manager.js &&
cp /proxy-patches/openai-to-cli.js $$DIST/adapter/openai-to-cli.js &&
cp /proxy-patches/cli-to-openai.js $$DIST/adapter/cli-to-openai.js &&
cp /proxy-patches/routes.js $$DIST/server/routes.js &&
claude-max-api"
volumes:
- ~/.claude:/root/.claude # Claude CLI Auth (Credentials in /root/.claude/.credentials.json)
+309
View File
@@ -0,0 +1,309 @@
/**
* ARIA-patched API Route Handlers
*
* Erweiterung der npm-Version von claude-max-api-proxy:
* - Bei jedem Claude-CLI-`assistant`-Event mit tool_use-Block (Bash, Read,
* Edit, Grep, ) wird ein HTTP-POST an die Bridge gefeuert
* (ARIA_TOOL_HOOK_URL, default http://aria-bridge:8090/internal/agent-activity).
* Bridge spiegelt das als RVS `agent_activity` an App+Diagnostic
* Gedanken-Stream zeigt live was ARIA gerade tool-maessig macht.
* - Fire-and-forget, fail-open. Wenn die Bridge nicht antwortet, bricht
* der Brain-Call NICHT ab.
*
* Wird zur Container-Startzeit ueber die npm-Version geschrieben
* (siehe docker-compose.yml proxy-Block).
*/
import { v4 as uuidv4 } from "uuid";
import http from "http";
import { ClaudeSubprocess } from "../subprocess/manager.js";
import { openaiToCli } from "../adapter/openai-to-cli.js";
import { cliResultToOpenai, createDoneChunk, } from "../adapter/cli-to-openai.js";
const TOOL_HOOK_URL = process.env.ARIA_TOOL_HOOK_URL
|| "http://aria-bridge:8090/internal/agent-activity";
/**
* Pusht einen Tool-Use-Event an die Bridge. Fire-and-forget keine Awaits,
* keine Fehler nach oben. Logged Fehler still.
*/
function _emitToolEvent(toolName) {
if (!toolName) return;
try {
const u = new URL(TOOL_HOOK_URL);
const body = JSON.stringify({ tool: String(toolName) });
const req = http.request({
method: "POST",
hostname: u.hostname,
port: u.port || 80,
path: u.pathname,
headers: { "Content-Type": "application/json", "Content-Length": Buffer.byteLength(body) },
timeout: 2000,
}, (res) => { res.resume(); });
req.on("error", () => {});
req.on("timeout", () => req.destroy());
req.write(body);
req.end();
} catch (_) { /* niemals weiterwerfen */ }
}
/**
* Hookt die `assistant`-Events des Subprozesses. Jedes assistant-Message
* kann mehrere content-Bloecke haben tool_use-Bloecke pushen wir live.
*/
function _attachToolHook(subprocess) {
subprocess.on("assistant", (message) => {
try {
const blocks = message?.message?.content || [];
for (const b of blocks) {
if (b && b.type === "tool_use" && b.name) {
_emitToolEvent(b.name);
}
}
} catch (_) { /* fail-open */ }
});
}
/**
* Handle POST /v1/chat/completions
*
* Main endpoint for chat requests, supports both streaming and non-streaming
*/
export async function handleChatCompletions(req, res) {
const requestId = uuidv4().replace(/-/g, "").slice(0, 24);
const body = req.body;
const stream = body.stream === true;
try {
// Validate request
if (!body.messages || !Array.isArray(body.messages) || body.messages.length === 0) {
res.status(400).json({
error: {
message: "messages is required and must be a non-empty array",
type: "invalid_request_error",
code: "invalid_messages",
},
});
return;
}
// Convert to CLI input format
const cliInput = openaiToCli(body);
const subprocess = new ClaudeSubprocess();
// ARIA-Patch: Tool-Use-Events live an die Bridge weiterleiten.
// Greift fuer beide Branches (stream + non-stream).
_attachToolHook(subprocess);
if (stream) {
await handleStreamingResponse(req, res, subprocess, cliInput, requestId);
}
else {
await handleNonStreamingResponse(res, subprocess, cliInput, requestId);
}
}
catch (error) {
const message = error instanceof Error ? error.message : "Unknown error";
console.error("[handleChatCompletions] Error:", message);
if (!res.headersSent) {
res.status(500).json({
error: {
message,
type: "server_error",
code: null,
},
});
}
}
}
/**
* Handle streaming response (SSE)
*
* IMPORTANT: The Express req.on("close") event fires when the request body
* is fully received, NOT when the client disconnects. For SSE connections,
* we use res.on("close") to detect actual client disconnection.
*/
async function handleStreamingResponse(req, res, subprocess, cliInput, requestId) {
// Set SSE headers
res.setHeader("Content-Type", "text/event-stream");
res.setHeader("Cache-Control", "no-cache");
res.setHeader("Connection", "keep-alive");
res.setHeader("X-Request-Id", requestId);
// CRITICAL: Flush headers immediately to establish SSE connection
// Without this, headers are buffered and client times out waiting
res.flushHeaders();
// Send initial comment to confirm connection is alive
res.write(":ok\n\n");
return new Promise((resolve, reject) => {
let isFirst = true;
let lastModel = "claude-sonnet-4";
let isComplete = false;
// Handle actual client disconnect (response stream closed)
res.on("close", () => {
if (!isComplete) {
// Client disconnected before response completed - kill subprocess
subprocess.kill();
}
resolve();
});
// Handle streaming content deltas
subprocess.on("content_delta", (event) => {
const text = event.event.delta?.text || "";
if (text && !res.writableEnded) {
const chunk = {
id: `chatcmpl-${requestId}`,
object: "chat.completion.chunk",
created: Math.floor(Date.now() / 1000),
model: lastModel,
choices: [{
index: 0,
delta: {
role: isFirst ? "assistant" : undefined,
content: text,
},
finish_reason: null,
}],
};
res.write(`data: ${JSON.stringify(chunk)}\n\n`);
isFirst = false;
}
});
// Handle final assistant message (for model name)
subprocess.on("assistant", (message) => {
lastModel = message.message.model;
});
subprocess.on("result", (_result) => {
isComplete = true;
if (!res.writableEnded) {
// Send final done chunk with finish_reason
const doneChunk = createDoneChunk(requestId, lastModel);
res.write(`data: ${JSON.stringify(doneChunk)}\n\n`);
res.write("data: [DONE]\n\n");
res.end();
}
resolve();
});
subprocess.on("error", (error) => {
console.error("[Streaming] Error:", error.message);
if (!res.writableEnded) {
res.write(`data: ${JSON.stringify({
error: { message: error.message, type: "server_error", code: null },
})}\n\n`);
res.end();
}
resolve();
});
subprocess.on("close", (code) => {
// Subprocess exited - ensure response is closed
if (!res.writableEnded) {
if (code !== 0 && !isComplete) {
// Abnormal exit without result - send error
res.write(`data: ${JSON.stringify({
error: { message: `Process exited with code ${code}`, type: "server_error", code: null },
})}\n\n`);
}
res.write("data: [DONE]\n\n");
res.end();
}
resolve();
});
// Start the subprocess
subprocess.start(cliInput.prompt, {
model: cliInput.model,
sessionId: cliInput.sessionId,
}).catch((err) => {
console.error("[Streaming] Subprocess start error:", err);
reject(err);
});
});
}
/**
* Handle non-streaming response
*/
async function handleNonStreamingResponse(res, subprocess, cliInput, requestId) {
return new Promise((resolve) => {
let finalResult = null;
subprocess.on("result", (result) => {
finalResult = result;
});
subprocess.on("error", (error) => {
console.error("[NonStreaming] Error:", error.message);
res.status(500).json({
error: {
message: error.message,
type: "server_error",
code: null,
},
});
resolve();
});
subprocess.on("close", (code) => {
if (finalResult) {
res.json(cliResultToOpenai(finalResult, requestId));
}
else if (!res.headersSent) {
res.status(500).json({
error: {
message: `Claude CLI exited with code ${code} without response`,
type: "server_error",
code: null,
},
});
}
resolve();
});
// Start the subprocess
subprocess
.start(cliInput.prompt, {
model: cliInput.model,
sessionId: cliInput.sessionId,
})
.catch((error) => {
res.status(500).json({
error: {
message: error.message,
type: "server_error",
code: null,
},
});
resolve();
});
});
}
/**
* Handle GET /v1/models
*
* Returns available models
*/
export function handleModels(_req, res) {
res.json({
object: "list",
data: [
{
id: "claude-opus-4",
object: "model",
owned_by: "anthropic",
created: Math.floor(Date.now() / 1000),
},
{
id: "claude-sonnet-4",
object: "model",
owned_by: "anthropic",
created: Math.floor(Date.now() / 1000),
},
{
id: "claude-haiku-4",
object: "model",
owned_by: "anthropic",
created: Math.floor(Date.now() / 1000),
},
],
});
}
/**
* Handle GET /health
*
* Health check endpoint
*/
export function handleHealth(_req, res) {
res.json({
status: "ok",
provider: "claude-code-cli",
timestamp: new Date().toISOString(),
});
}
//# sourceMappingURL=routes.js.map