Files
platform/src/servers/api/chat/websocket.ts
T
pastilhasandClaude Opus 5 73b8111216 send images to opencode, which never needed the fork
B4 properly. The composer gate was the honest stopgap; this is the fix. opencode run takes
attachments with --file, so images work on the subprocess path we already use — the parity
doc had them down as phase 4, behind the serve migration, and they were not.

The bug was one omission: handleOpenCodeChat`s msg type had no images field, so the browser
sent them, the bubble rendered them, and they stopped at that signature. Nothing reported a
loss anywhere.

Attachments are paths, not inline data, so the sidecar spills each image to a temp file for
the length of the turn and removes it in settle — the same place every other per-turn
resource is released, so a killed or superseded turn cleans up too.

The load-bearing detail is `--` before the prompt: --file is an array option, so without the
separator the prompt is eaten as another filename and the turn dies with "File not found:"
followed by the entire message. Confirmed against the binary, and pinned by a test that
records argv from a stub.

list-models now reports each model own capability instead of a hardcoded false — opencode
publishes capabilities.input.image per model and nothing had ever read it. Defaults to false,
so a model that does not declare it keeps the affordance hidden.

Verified end to end: a red png sent over the chat socket to opencode/claude-sonnet-4-6 came
back "Red".

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2026-08-10 16:49:39 +01:00

849 lines
33 KiB
TypeScript

import type { ServerWebSocket } from 'bun';
import { randomUUID } from 'crypto';
import type {
ClientMessage,
ServerMessage,
Message,
PromptImage,
RunningTask,
TurnMessage,
UserSession,
} from './types';
import { sessionManager } from './session-manager';
import { sendClaudeCodeStreaming } from '@@/channels/send-claude-code';
import { sendOpenCodeStreaming } from '@@/channels/send-opencode';
import { ensureGeneralChatSessionsCwd } from './claude-sessions';
import * as sidecar from '@@/sidecar-registry';
import { join } from 'path';
import { getOwnerHomeDir, getEmailAccountsDir } from '../../../servers/data-path';
import { getUserSettings, getEmailAccounts, getChatEventsSince, appendChatEvent } from 'officerdb';
import { mkdirSync } from 'node:fs';
import { logger } from './logger';
// Default model when no user preference is set
const DEFAULT_MODEL = 'claude-code';
// Harness selection: the `claude-code` provider runs through the Claude sidecar; every other provider
// (opencode/anthropic/openai/… — all `providerID/modelID` ids) runs through the OpenCode server.
const isClaudeModel = (model: string): boolean => model.startsWith('claude-code');
async function getUserDefaultModel(userId: number): Promise<string | null> {
try {
const settings = await getUserSettings(userId);
const chat = settings?.chat as Record<string, unknown> | undefined;
return (chat?.defaultModel as string) || null;
} catch (err) {
logger.error('Failed to read user settings for default model', { userId, error: String(err) });
}
return null;
}
type WSData = {
userId: number;
email: string;
username: string;
provider: string;
};
const IDLE_TIMEOUT_MS = 60 * 60 * 1000; // 1 hour
const resolveCwd = (email: string, cwd?: string) => {
const root = getOwnerHomeDir(email);
if (!cwd || cwd === '~') return root;
if (cwd.startsWith('~/')) return join(root, cwd.slice(2));
// The server owner is the only account — absolute paths are theirs to use.
if (cwd.startsWith('/')) return cwd;
return join(root, cwd);
};
export const resolveBaseCwd = (email: string, cwd?: string) => resolveCwd(email, cwd);
// The email chat runs from the selected account's storage dir:
// DATA_PATH/<owner>/email_accounts/<accountEmail>
// `accountEmail` will come from the account selector (msg.contextId) later; for now default to the
// owner's first enabled account. Falls back to the email_accounts root if there are no accounts.
async function resolveEmailCwd(userId: number, ownerEmail: string, accountEmail?: string): Promise<string> {
let account = accountEmail?.trim();
if (!account) {
try {
const accounts = await getEmailAccounts(userId);
account = (accounts.find((a) => a.enabled) ?? accounts[0])?.email;
} catch (err) {
logger.error('Failed to resolve email account for chat cwd', { userId, error: String(err) });
}
}
const dir = account ? join(getEmailAccountsDir(ownerEmail), account) : getEmailAccountsDir(ownerEmail);
mkdirSync(dir, { recursive: true });
return dir;
}
// The working directory a chat turn runs in, by context: email → the account dir; /chat → a chosen
// pwd or the default general_chat_sessions dir; everything else (browser/project/dashboard) → the given cwd.
async function resolveChatCwd(
msg: { context?: string; contextId?: string; cwd?: string },
email: string,
userId: number,
): Promise<string> {
if (msg.context === 'email') return resolveEmailCwd(userId, email, msg.contextId);
if (msg.context === 'chat') return msg.cwd?.trim() ? resolveCwd(email, msg.cwd) : ensureGeneralChatSessionsCwd(email);
return resolveCwd(email, msg.cwd);
}
const wsToSessionMap = new WeakMap<any, string>();
// Per-connection heartbeat. Bun closes a WS idle for `idleTimeout` (60s), and its timer only resets
// on frames *received* from the client — but during a chat turn the client only receives. So we ping
// each connection every 25s; the client auto-pongs at the protocol level, which resets Bun's timer
// (and keeps reverse proxies happy). Genuinely dead sockets still time out (no pong).
const pingTimers = new WeakMap<ServerWebSocket<WSData>, ReturnType<typeof setInterval>>();
const PING_INTERVAL_MS = 25_000;
function sendToClient(ws: ServerWebSocket<WSData> | null, msg: ServerMessage, seq?: number): void {
if (ws?.readyState === 1) {
ws.send(JSON.stringify(seq === undefined ? msg : { ...msg, seq }));
}
}
// Nothing in officer writes to chat_session_events any more. Both harnesses commit their own turn
// output in the sidecar that produced it (`sidecar/claude/session-log.ts`), which is the whole point:
// the durable record does not travel over the socket between the two processes, so officer can restart
// mid-turn without losing it. Officer reads the table on `resume` (getChatEventsSince) and relays what
// the sidecars send. The old `emitToSession` used to live here.
export async function open(ws: ServerWebSocket<WSData>): Promise<void> {
const timer = setInterval(() => {
try {
ws.ping(); // client auto-pongs → resets Bun's idleTimeout
} catch {
/* socket already gone */
}
}, PING_INTERVAL_MS);
pingTimers.set(ws, timer);
}
export function message(ws: ServerWebSocket<WSData>, raw: string | Buffer): void {
const data = typeof raw === 'string' ? raw : raw.toString();
(async () => {
try {
const clientMsg = JSON.parse(data) as ClientMessage;
if (clientMsg.type === 'chat') {
await handleChat(ws, clientMsg);
} else if (clientMsg.type === 'resume') {
await handleResume(ws, clientMsg);
} else if (clientMsg.type === 'stop') {
await handleStop(ws);
} else if (clientMsg.type === 'disconnect') {
await handleDisconnect(ws);
} else if (clientMsg.type === 'resume-cursor') {
await handleResumeCursor(ws, clientMsg);
} else if (clientMsg.type === 'attach') {
await handleAttach(ws, clientMsg);
}
} catch (err) {
logger.error('Error handling WebSocket message', { email: ws.data.email, error: String(err) });
sendToClient(ws, { type: 'error', message: 'Failed to process message' });
}
})();
}
export function close(ws: ServerWebSocket<WSData>): void {
const timer = pingTimers.get(ws);
if (timer) {
clearInterval(timer);
pingTimers.delete(ws);
}
const sessionId = wsToSessionMap.get(ws);
if (sessionId) {
sessionManager.detachWs(sessionId);
sessionManager.setIdleTimeout(sessionId, IDLE_TIMEOUT_MS);
}
}
// ── Claude Code: relay, don't rebuild ──
// The agent sidecar owns Claude's turn output end to end — it translates the parser stream, commits each
// message to chat_session_events and hands us a finished message plus its cursor id. Officer relays it.
// That is what makes a restart survivable: the durable record no longer travels over the socket between
// the two processes, so if this one is down the output is already written and the client replays it.
//
// Officer keeps only the in-memory transcript, which exists to answer a `resume` with sync:messages —
// Claude's own transcript is the real record — so it is folded from the same messages, not rebuilt.
function foldIntoSession(session: UserSession, msg: TurnMessage, model: string): void {
switch (msg.type) {
case 'assistant:delta':
// A subagent's deltas are not the main agent typing; appending them here spliced its sentences into
// whatever the agent you are talking to was mid-way through saying.
if (!msg.parentToolUseId) session.streamBuffer += msg.text;
break;
case 'assistant:text':
session.messages.push({
id: randomUUID(),
timestamp: Date.now(),
role: 'assistant',
text: msg.text,
model,
parentToolUseId: msg.parentToolUseId,
});
session.meta.messageCount += 1;
// Only the main agent's own stream feeds the buffer a resume replays as `streamingText`.
if (!msg.parentToolUseId) session.streamBuffer = '';
break;
case 'tool:start':
session.messages.push({
id: randomUUID(),
timestamp: Date.now(),
role: 'tool',
toolCallId: msg.toolCallId,
toolName: msg.toolName,
toolInput: msg.toolInput,
parentToolUseId: msg.parentToolUseId,
});
session.meta.messageCount += 1;
break;
case 'tool:result':
for (let i = session.messages.length - 1; i >= 0; i--) {
const m = session.messages[i]!;
if (m.role === 'tool' && m.toolCallId === msg.toolCallId) {
m.output = msg.output;
m.isError = msg.isError;
break;
}
}
break;
case 'result': {
session.isGenerating = false;
session.meta.cost.inputTokens += msg.cost.inputTokens;
session.meta.cost.outputTokens += msg.cost.outputTokens;
session.meta.cost.totalUSD += msg.cost.totalUSD;
session.meta.updatedAt = Date.now();
// The turn's cost belongs to the assistant message it paid for (as it did when officer built these).
// The turn's cost is the main agent's, so skip past any subagent tail.
const last = [...session.messages].reverse().find((m) => !m.parentToolUseId);
if (last?.role === 'assistant' && !last.cost) last.cost = msg.cost;
break;
}
case 'error':
case 'stopped':
session.isGenerating = false;
break;
}
}
// Shared by both harnesses now: whichever sidecar ran the turn has already translated and committed it.
function createMessageHandler(sessionId: string, model: string) {
return (msg: TurnMessage, seq?: number): void => {
const session = sessionManager.getSession(sessionId);
if (!session) return;
foldIntoSession(session, msg, model);
sendToClient(session.ws as ServerWebSocket<WSData> | null, msg, seq);
};
}
async function handleChat(
ws: ServerWebSocket<WSData>,
msg: {
prompt: string;
displayText?: string;
sessionId?: string;
model?: string;
cwd?: string;
cwdRoot?: string;
groupSlug?: string;
attachmentIds?: string[];
images?: PromptImage[];
thinking?: string;
context?: string;
contextId?: string;
resumeSummary?: string;
resumeSessionId?: string;
},
): Promise<void> {
const { userId } = ws.data;
const sessionId = msg.sessionId || randomUUID();
// Prepend resume summary to the prompt if present
const prompt = msg.resumeSummary
? `Here is a summary of a previous conversation to continue from:\n\n${msg.resumeSummary}\n\n---\n\nUser's new message: ${msg.prompt}`
: msg.prompt;
// Use provided model, or fall back to user default, or the system default.
const model = msg.model || (await getUserDefaultModel(userId)) || DEFAULT_MODEL;
logger.info('Model selected for chat', { sessionId, model, clientModel: msg.model || null });
// Route by harness: claude-code → Claude sidecar; anything else → OpenCode server.
return isClaudeModel(model)
? handleClaudeCodeChat(ws, sessionId, model, msg, prompt)
: handleOpenCodeChat(ws, sessionId, model, msg, prompt);
}
async function handleClaudeCodeChat(
ws: ServerWebSocket<WSData>,
sessionId: string,
model: string,
msg: {
prompt: string;
displayText?: string;
groupSlug?: string;
context?: string;
contextId?: string;
cwd?: string;
cwdRoot?: string;
images?: PromptImage[];
resumeSessionId?: string;
},
effectivePrompt: string,
): Promise<void> {
const { email, username, userId } = ws.data;
const cwd = await resolveChatCwd(msg, email, userId);
const groupSlug = msg.groupSlug || null;
const session = sessionManager.getOrCreate(sessionId, email, cwd, model, groupSlug, msg.context, msg.contextId);
session.userId = userId;
sessionManager.attachWs(sessionId, ws);
wsToSessionMap.set(ws as any, sessionId);
sendToClient(ws, {
type: 'session:init',
sessionId,
model,
cwd,
context: session.meta.context,
contextId: session.meta.contextId,
});
// Add user message to session
const userMsg: Message = {
id: randomUUID(),
timestamp: Date.now(),
role: 'user',
text: msg.prompt,
};
session.messages.push(userMsg);
session.meta.messageCount += 1;
session.meta.updatedAt = Date.now();
if (!session.meta.title) {
session.meta.title = (msg.displayText ?? msg.prompt).slice(0, 100);
}
session.isGenerating = true;
const onMessage = createMessageHandler(sessionId, model);
try {
if (!session._claudeKill) {
// First turn of this session: open the persistent session + a SESSION-scoped event subscription
// (survives turn-end so background task:notifications keep flowing). `kill` tears both down for an
// explicit disconnect; `detach` drops only the listener, which is what the idle GC uses so an
// absent browser stops taking a live agent with it.
const handle = await sendClaudeCodeStreaming({
userId,
email,
username,
prompt: effectivePrompt,
images: msg.images,
sessionKey: sessionId,
cwd,
model,
resumeSessionId: msg.resumeSessionId,
onMessage,
});
session.piProcess = sessionId as any;
session._claudeKill = handle.kill;
session._sidecarUnsub = handle.detach;
} else {
// Session already live: push this turn onto the existing persistent session (no new subscription).
await sidecar.spawnClaudeStreaming({
userId,
email,
username,
prompt: effectivePrompt,
images: msg.images,
sessionKey: sessionId,
cwd,
model,
resumeSessionId: msg.resumeSessionId,
});
}
} catch (err) {
logger.error('Failed to start Claude Code streaming', { sessionId, error: String(err) });
sendToClient(ws, { type: 'error', message: 'Failed to start Claude Code' });
session.isGenerating = false;
}
}
async function handleOpenCodeChat(
ws: ServerWebSocket<WSData>,
sessionId: string,
model: string,
msg: {
prompt: string;
displayText?: string;
groupSlug?: string;
context?: string;
contextId?: string;
cwd?: string;
cwdRoot?: string;
resumeSessionId?: string;
// The whole of B4 lived in this omission. The browser sent images, the bubble rendered them, and
// they stopped at this signature — so they were never passed on and never reached the model, with
// nothing anywhere reporting a loss.
images?: PromptImage[];
},
effectivePrompt: string,
): Promise<void> {
const { email, username, userId } = ws.data;
const cwd = await resolveChatCwd(msg, email, userId);
const groupSlug = msg.groupSlug || null;
const session = sessionManager.getOrCreate(sessionId, email, cwd, model, groupSlug, msg.context, msg.contextId);
session.userId = userId;
sessionManager.attachWs(sessionId, ws);
wsToSessionMap.set(ws as any, sessionId);
sendToClient(ws, {
type: 'session:init',
sessionId,
model,
cwd,
context: session.meta.context,
contextId: session.meta.contextId,
});
const userMsg: Message = {
id: randomUUID(),
timestamp: Date.now(),
role: 'user',
text: msg.prompt,
};
session.messages.push(userMsg);
session.meta.messageCount += 1;
session.meta.updatedAt = Date.now();
if (!session.meta.title) {
session.meta.title = (msg.displayText ?? msg.prompt).slice(0, 100);
}
session.isGenerating = true;
const onMessage = createMessageHandler(sessionId, model);
try {
// Drop the PREVIOUS turn's listener before opening the next one.
//
// This deliberately does not mirror the Claude guard above. Claude keeps one persistent session and
// skips re-subscribing; OpenCode runs a fresh `opencode run` subprocess per turn, so a new
// subscription each time is correct. What was wrong is that the old handle was overwritten without
// being detached, leaving the previous session-scoped listener attached — so every turn after the
// first delivered doubled, tripled, and so on, for any termination that is not result/error/stopped.
session._sidecarUnsub?.();
session._sidecarUnsub = undefined;
const handle = await sendOpenCodeStreaming({
userId,
email,
username,
prompt: effectivePrompt,
sessionKey: sessionId,
cwd,
model,
resumeSessionId: msg.resumeSessionId,
images: msg.images,
onMessage,
});
// Store the abort handle so handleStop can end the turn (OpenCode is aborted via this handle).
session.piProcess = sessionId as any;
session._claudeKill = handle.kill;
session._sidecarUnsub = handle.detach;
} catch (err) {
logger.error('Failed to start OpenCode streaming', { sessionId, error: String(err) });
sendToClient(ws, { type: 'error', message: 'Failed to start OpenCode' });
session.isGenerating = false;
}
}
async function handleResume(
ws: ServerWebSocket<WSData>,
msg: { sessionId: string; cwd?: string; cwdRoot?: string },
): Promise<void> {
const { sessionId } = msg;
try {
const session = sessionManager.getSession(sessionId);
if (!session) {
// Sessions live in memory for the connection's lifetime; there's no disk store to reload from.
sendToClient(ws, { type: 'error', message: 'Session not found', errorCode: 'SESSION_NOT_FOUND' });
return;
}
sessionManager.attachWs(sessionId, ws);
wsToSessionMap.set(ws as any, sessionId);
sendToClient(ws, {
type: 'session:init',
sessionId,
model: session.model,
cwd: session.cwd,
context: session.meta.context,
contextId: session.meta.contextId,
});
// Claude resumes lazily: the next chat prompt re-attaches via `--resume <sessionKey>`, so there's
// no long-lived process to spawn here — just replay the stored transcript to the client.
sendToClient(ws, {
type: 'sync:messages',
sessionId,
messages: session.messages,
isGenerating: session.isGenerating,
streamingText: session.streamBuffer,
});
logger.info('Session resumed successfully', { sessionId, messageCount: session.messages.length });
} catch (err) {
logger.error('Unexpected error in handleResume', { sessionId, error: String(err) });
sendToClient(ws, { type: 'error', message: 'Failed to resume session' });
}
}
async function handleStop(ws: ServerWebSocket<WSData>): Promise<void> {
const sessionId = wsToSessionMap.get(ws);
if (sessionId) {
const session = sessionManager.getSession(sessionId);
if (session?.piProcess) {
try {
if (isClaudeModel(session.model)) {
// Interrupt the current turn but KEEP the persistent session alive (background tasks + the
// warm worker survive). Full teardown is 'disconnect' → deleteSession → _claudeKill.
void sidecar.interruptClaude(sessionId);
logger.info('Interrupted Claude Code turn via sidecar (session stays warm)', { sessionId });
} else {
session._claudeKill?.(); // OpenCode: abort the turn via the stored handle
logger.info('Aborted OpenCode turn', { sessionId });
}
session.isGenerating = false;
} catch (err) {
logger.error('Failed to stop process', { sessionId, error: String(err) });
}
}
}
sendToClient(ws, { type: 'stopped' });
}
// Tear down the whole session (not just the current turn): deleteSession fires _claudeKill (kills any
// in-flight Claude/OpenCode turn) + _sidecarUnsub, clears the idle timer, and drops the session from the
// manager's maps. The WS stays open so the client can immediately start a fresh session.
async function handleDisconnect(ws: ServerWebSocket<WSData>): Promise<void> {
const sessionId = wsToSessionMap.get(ws);
if (sessionId) {
try {
sessionManager.deleteSession(sessionId);
wsToSessionMap.delete(ws as any);
logger.info('Disconnected chat session', { sessionId });
} catch (err) {
logger.error('Failed to disconnect session', { sessionId, error: String(err) });
}
}
sendToClient(ws, { type: 'disconnected' });
}
/**
* Re-adopt a session whose in-memory record died with the process that held it.
*
* The sidecars are PM2 peers, so `pm2 restart officer` does not touch a running turn: the agent keeps
* generating and keeps committing to chat_session_events. What the restart destroys is purely this
* process's binding to it — the session record and, critically, the session-scoped subscription that
* relays the sidecar's events to the browser. Re-creating the record is not enough on its own; without
* the subscription the client reconnects, receives its replay, and then goes silent for the rest of the
* turn, which is indistinguishable from the agent having died.
*
* Nothing is spawned here. The subscription is a local event-bus filter, so adopting a session that is
* NOT in fact still live upstream costs a listener that never fires and a record the idle GC collects.
*/
function adoptOrphanedSession(ws: ServerWebSocket<WSData>, sessionId: string, model: string, cwd: string): UserSession {
const { email, userId } = ws.data;
const session = sessionManager.getOrCreate(sessionId, email, cwd, model, null);
session.userId = userId;
session.piProcess = sessionId as any;
const onMessage = createMessageHandler(sessionId, model);
const unsubClaude = sidecar.onClaudeMessage((key, msg, seq) => {
if (key === sessionId) onMessage(msg, seq);
});
const unsubOpenCode = sidecar.onOpenCodeMessage((key, msg, seq) => {
if (key === sessionId) onMessage(msg, seq);
});
const unsub = () => {
unsubClaude();
unsubOpenCode();
};
// Must be set, and not only for teardown: handleChat treats an absent `_claudeKill` as "first turn of
// this session" and opens a *second* subscription, which would then deliver every message twice.
session._claudeKill = () => {
if (isClaudeModel(model)) sidecar.killClaude(sessionId);
else sidecar.killOpenCode(sessionId);
unsub();
};
// An adopted session can idle out and be released like any other, and releasing detaches through this
// field alone. Leaving it unset would drop the record while the listener stayed subscribed — a leak
// that grows by one every time a browser adopts a session and then goes away.
session._sidecarUnsub = unsub;
logger.info('Adopted orphaned chat session after restart', { sessionId, model });
return session;
}
// Reconnect: re-bind this socket to the (possibly still-live) session and replay every durable event
// queued since the client's cursor — so a brief disconnect never loses turn output or a background
// task:notification. attachWs cancels the pending idle-GC.
async function handleResumeCursor(
ws: ServerWebSocket<WSData>,
msg: { sessionId: string; cursor: number; model?: string; cwd?: string; generating?: boolean },
): Promise<void> {
const { sessionId, cursor } = msg;
const known = sessionManager.getSession(sessionId);
const decision = decideResume(known?.model, msg.model);
const model = decision.kind === 'replay-only' ? null : decision.model;
if (decision.kind === 'adopt') {
adoptOrphanedSession(ws, sessionId, decision.model, msg.cwd ?? '');
} else if (decision.kind === 'replay-only') {
// Replay the durable log and stop there. Not adopting costs a live re-subscription; adopting on a
// guess cost correctness — see decideResume.
logger.warn('resume-cursor names a session this process does not know, and no model; replaying only', {
sessionId,
});
}
sessionManager.attachWs(sessionId, ws); // a no-op when adoption was skipped
wsToSessionMap.set(ws as any, sessionId);
try {
const events = await getChatEventsSince(sessionId, cursor ?? 0);
for (const { id, event } of events) {
sendToClient(ws, event as ServerMessage, id);
}
} catch (err) {
logger.error('Failed to replay chat events on resume-cursor', { sessionId, error: String(err) });
}
// Only ask when the harness is actually known. The check writes a DURABLE row, so a wrong answer here
// is permanent — silence is the safe failure.
if (msg.generating && model) await endTurnIfAgentIsGone(ws, sessionId, model);
}
export type ResumeDecision =
| { kind: 'known'; model: string }
| { kind: 'adopt'; model: string }
| { kind: 'replay-only' };
/**
* Which harness a resuming session belongs to, and what that permits.
*
* This used to be `msg.model || DEFAULT_MODEL`, and `DEFAULT_MODEL` is `claude-code`. So a resume-cursor
* that omitted `model` declared every session — OpenCode ones included — to be Claude, with three
* consequences that all read as something else:
*
* - **Adopted into the wrong harness.** `adoptOrphanedSession` subscribes to the sidecar bus for that
* model and pins `session.model` for the rest of its life, so an OpenCode turn's output never
* arrived, and stopping it called `killClaude` on a key that sidecar had never held — a dead stop
* button, silently.
* - **A durable false `cut-off`.** `endTurnIfAgentIsGone` asked the Claude sidecar whether it was
* generating, was told `false` because it had never heard of the session, and wrote "the agent went
* away" into a turn that was running normally. It survives reload, which is the whole point of
* writing it durably, and is therefore unrecoverable from the UI.
* - Masked, never fixed, by the client always happening to send `model` next to `sessionId`.
*
* Two rules. **The server's own record beats the client's claim** — a session in memory already knows its
* harness, and letting a socket re-declare it is how the wrong sidecar gets a session in the first place.
* **An unknown harness stays unknown**: no adoption, no cut-off check, just the replay. Defaulting is
* what made a guess indistinguishable from knowledge.
*/
export function decideResume(knownModel: string | undefined, claimedModel: string | undefined): ResumeDecision {
if (knownModel) return { kind: 'known', model: knownModel };
if (claimedModel) return { kind: 'adopt', model: claimedModel };
return { kind: 'replay-only' };
}
/**
* Which background tasks are still outstanding, by replaying the durable log against itself.
*
* A task's whole life is two events — `task:started` and, eventually, `task:notification` with a terminal
* status — so started-minus-notified is the answer, and a Map keyed by taskId keeps the last word on each.
* There is no third event: a task the agent abandoned without notifying stays here until the log is pruned,
* which is the honest reading of the record rather than a bug to paper over.
*/
function collectRunningTasks(events: ServerMessage[]): RunningTask[] {
const running = new Map<string, RunningTask>();
for (const event of events) {
if (event.type === 'task:started') {
running.set(event.taskId, {
taskId: event.taskId,
description: event.description,
taskType: event.taskType,
});
} else if (event.type === 'task:notification') {
running.delete(event.taskId);
}
}
return [...running.values()];
}
/**
* Re-bind a socket that knows only Claude's transcript uuid.
*
* This is the refresh case, and until now it was the hole in an otherwise complete reconnect path. Every
* piece of the machinery already existed — the session survives a dropped socket, the agent keeps
* generating into it, `close` only detaches and arms an hour-long idle timer — but the browser came back
* having forgotten officer's session id, so `resume-cursor` could never fire and the output simply stopped
* arriving. The uuid in the URL is the one identifier a refresh cannot destroy; the agent's on-disk map
* turns it back into the key everything else here is written in terms of.
*
* Deliberately hands over only the live turn, never the transcript — see `sync:live`.
*/
async function handleAttach(ws: ServerWebSocket<WSData>, msg: { claudeSessionId: string }): Promise<void> {
const { claudeSessionId } = msg;
if (!claudeSessionId) return;
const sessionId = await sidecar.findClaudeSessionKey(claudeSessionId);
if (!sessionId) {
// No agent, or a transcript it has never run. Nothing is wrong: an ordinary finished conversation
// opened from history lands here every time. Stay silent and leave the socket as it was — the next
// `chat` mints a session in the usual way.
logger.info('Attach found no live session for transcript', { claudeSessionId });
return;
}
// An officer restart takes the in-memory session with it while the agent carries on, so the key can
// resolve to a session this process has never heard of. Adopting re-subscribes it to the sidecar's bus,
// which is what makes the rest of the turn arrive.
// `DEFAULT_MODEL` is right here and wrong in resume-cursor, which is worth being explicit about since
// it was just removed there: this path reached `sessionId` by asking the CLAUDE sidecar to resolve a
// `claudeSessionId`, so the harness is not a guess — only Claude could have answered. Attach is
// Claude-only by construction. If an OpenCode reattach verb ever lands, this stops being safe.
const existing = sessionManager.getSession(sessionId);
const session = existing ?? adoptOrphanedSession(ws, sessionId, DEFAULT_MODEL, '');
sessionManager.attachWs(sessionId, ws);
wsToSessionMap.set(ws as any, sessionId);
// The client learns officer's key here, so any *later* drop of this socket goes down the existing
// cursor-replay path instead of coming back through attach.
sendToClient(ws, {
type: 'session:init',
sessionId,
model: session.model,
cwd: session.cwd,
context: session.meta.context,
contextId: session.meta.contextId,
});
// `isGenerating` is officer's own belief and is only as good as this process's memory of the turn. For
// an adopted session it is a fresh record's default, so ask the agent — the same question, and for the
// same reason, as `endTurnIfAgentIsGone`.
const isGenerating = existing ? session.isGenerating : await sidecar.isClaudeGenerating(sessionId);
session.isGenerating = isGenerating;
// One read serves both answers: the head of the log is the cursor, and folding the whole log gives the
// tasks still outstanding. Reading it all is affordable because attach happens once per socket and this
// server has one user; a `getLastChatEventSeq` would only have saved a second round trip.
let cursor = 0;
let runningTasks: RunningTask[] = [];
try {
const events = await getChatEventsSince(sessionId, 0);
cursor = events.at(-1)?.id ?? 0;
runningTasks = collectRunningTasks(events.map((e) => e.event as ServerMessage));
} catch (err) {
logger.error('Failed to read the durable log on attach', { sessionId, error: String(err) });
}
sendToClient(ws, {
type: 'sync:live',
sessionId,
isGenerating,
cursor,
runningTasks,
// Whatever the agent had typed but not yet finished as a message. The transcript on disk cannot
// supply it — the harness writes an assistant message only once it is complete — so this is the one
// piece of the turn a refresh would otherwise genuinely lose.
streamingText: session.streamBuffer,
});
logger.info('Attached socket to live session by transcript id', {
sessionId,
claudeSessionId,
isGenerating,
recoveredTasks: runningTasks.length,
});
}
/**
* The client came back still believing a turn is running. Check whether it is, and if it isn't, say so.
*
* A restart of the agent sidecar takes its persistent sessions with it, and nothing downstream notices:
* the turn simply stops emitting. The browser's socket is fine, the conversation looks alive, and the
* spinner runs forever — a refresh doesn't help either, because there is no ending in the transcript to
* read. This is the one moment we can catch it, so the answer is written durably: a reload after this
* shows the same explanation rather than a conversation that trails off mid-tool-call.
*
* Only the claude harness is asked. OpenCode runs a turn per invocation and has no equivalent question,
* so its sessions are left alone rather than guessed at.
*/
async function endTurnIfAgentIsGone(
ws: ServerWebSocket<WSData> | null,
sessionId: string,
model: string,
): Promise<void> {
if (!isClaudeModel(model)) return;
if (await sidecar.isClaudeGenerating(sessionId)) return;
const session = sessionManager.getSession(sessionId);
if (session) session.isGenerating = false;
const event: ServerMessage = { type: 'cut-off' };
try {
const seq = await appendChatEvent(sessionId, event);
sendToClient(ws, event, seq);
} catch (err) {
// Still tell this client — an un-replayable explanation beats a spinner that never stops.
logger.error('Failed to persist cut-off notice', { sessionId, error: String(err) });
sendToClient(ws, event);
}
logger.info('Ended a turn whose agent had gone', { sessionId });
}
// The other half of the same problem: the agent restarts while the browser sits there with a healthy
// socket, so nothing ever reconnects and nothing ever asks. A fresh agent process means every turn we
// still believe is running belongs to a process that no longer exists. On a fresh officer this loop is
// empty — it has no sessions yet — which is exactly right, because that case is the reconnect's to catch.
sidecar.onClaudeSidecarStarted(() => {
for (const session of sessionManager.getAllSessions()) {
if (!session.isGenerating) continue;
void endTurnIfAgentIsGone(session.ws as ServerWebSocket<WSData> | null, session.sessionId, session.model);
}
});
export const chatWebsocket = {
open,
message,
close,
drain() {},
};