@evident-ai/cli 3.1.1-dev.716aea1 → 3.1.1-dev.7a5732a

package/dist/index.js

@@ -499,6 +499,12 @@ function log(level, event, fields) {
     );
   }
 }
+function errorFields(err) {
+  if (err instanceof Error) {
+    return { error: err.message, error_name: err.name };
+  }
+  return { error: String(err) };
+}
 function stripQuery(url) {
   try {
     return new URL(url).pathname;
@@ -1267,6 +1273,16 @@ async function buildFileParts(attachments, capable) {
       );
       dataUrl = null;
     }
+    if (dataUrl !== null && typeof dataUrl === "object") {
+      outcomes.push({
+        index: a.index,
+        mime: a.mime,
+        filename: a.filename,
+        status: "failed",
+        reason: "needs_reauth"
+      });
+      continue;
+    }
     if (dataUrl == null) {
       outcomes.push({ index: a.index, mime: a.mime, filename: a.filename, status: "failed" });
       continue;
@@ -1684,10 +1700,11 @@ var StreamForwarder = class {
    * Abort every in-flight stream (e.g. on WebSocket close).
    */
   abortAll() {
-    for (const stream of this.inflight.values()) {
+    for (const [sid, stream] of this.inflight.entries()) {
       try {
         stream.abort();
-      } catch {
+      } catch (err) {
+        log("error", "forwarder_abort_failed", { sid, ...errorFields(err) });
       }
     }
     this.inflight.clear();
@@ -1960,7 +1977,11 @@ var RunnerConnection = class {
     if (this.connection) {
       try {
         this.connection.close();
-      } catch {
+      } catch (err) {
+        log("error", "runner_connection_close_failed", {
+          agent_id: this.resolvedAgentId,
+          ...errorFields(err)
+        });
       }
       this.connection = null;
     }
@@ -2041,6 +2062,7 @@ var DEFAULT_STUCK_QUEUED_MS = 6e4;
 var HEARTBEAT_MS = 6e4;
 var ABSOLUTE_MAX_PROCESSING_MS = 6 * 60 * 60 * 1e3;
 var POLL_MISS_GRACE_MS = HEARTBEAT_MS;
+var MAX_SUPERSEDED_CONVERSATIONS = 256;
 var ChannelAuthError = class extends Error {
   constructor(message) {
     super(message);
@@ -2079,6 +2101,34 @@ var ChannelDriver = class _ChannelDriver {
   now;
   /** Cache of conversationId → opencode sessionId. */
   sessions = /* @__PURE__ */ new Map();
+  /**
+   * conversationId → the opencode session this runner has ABANDONED as that
+   * conversation's binding (#553), after a genuine (`sessionExists === true`)
+   * dispatch failure: the session still exists but is wedged, so #485's self-heal
+   * must bind a fresh one.
+   *
+   * Dropping the local binding + clearing the server row is not enough on its own:
+   * a SIBLING message dispatched earlier in the same drain is still in-flight under
+   * the same session, and its watcher's routine status writes carry
+   * `opencode_session_id`, RESURRECTING the wedged id server-side after the clear —
+   * and `ensureSession`'s persisted-id fallback then reuses it, defeating the
+   * self-heal. This map makes the runner authoritative instead of racing those
+   * writes: *`ensureSession` never reuses an abandoned id for that conversation,
+   * whatever the server row says* — which holds even when the resurrecting write
+   * is one we deliberately keep (see `markDone`).
+   *
+   * Bounded by construction, on both axes: keyed by CONVERSATION, so N failures on
+   * one conversation hold ONE entry (the newest abandonment replaces the older), and
+   * hard-capped at `MAX_SUPERSEDED_CONVERSATIONS` with FIFO eviction. Only the
+   * NEWEST abandoned id per conversation is guarded: after a second abandonment a
+   * late sibling of the FIRST session can write that id back and `ensureSession`
+   * will reuse it — costing ONE repeat failure, which re-supersedes it. Deliberately
+   * NOT dropped when the session's watcher tears down: `markDone` still writes the
+   * abandoned id back (it must, or the reply is lost), so the guard has to outlive
+   * the turn that resurrects it. In-memory only — a restart forgets it, at the same
+   * bounded cost.
+   */
+  supersededSessions = /* @__PURE__ */ new Map();
   /**
    * Per-opencode-session dispatch lock (Task 2.1a). `sendPromptAsync` is no
    * longer idempotent (no caller-supplied `messageID`), and its read-back picks
@@ -2387,10 +2437,15 @@ var ChannelDriver = class _ChannelDriver {
    * @returns the count of messages NEWLY dispatched (not already in-flight).
    */
   async processConversation(conv) {
-    const sessionId = await this.ensureSession(conv);
+    const { sessionId, refusedSessionId } = await this.ensureSession(conv);
     const messages = await this.getPendingMessages(conv.id);
     let dispatched = 0;
     let skippedAlreadyDispatched = 0;
+    if (refusedSessionId && messages.length > 0) {
+      void this.postSignal(conv.id, messages[0].id, "session_superseded", {
+        superseded_session_id: refusedSessionId
+      });
+    }
     for (const message of messages) {
       if (this.stopped) break;
       if (this.dispatched.has(message.id)) {
@@ -2417,7 +2472,8 @@ var ChannelDriver = class _ChannelDriver {
       } catch (err) {
         if (err instanceof ChannelAuthError) throw err;
         this.dispatched.delete(message.id);
-        if (await sessionExists(this.port, sessionId) === false) {
+        const exists = await sessionExists(this.port, sessionId);
+        if (exists === false) {
           this.sessions.delete(conv.id);
           this.log({
             level: "warn",
@@ -2427,15 +2483,39 @@ var ChannelDriver = class _ChannelDriver {
           });
           break;
         }
-        await this.markFailed(conv.id, message.id).catch(() => {
+        if (exists === null) {
+          this.log({
+            level: "warn",
+            message: `Message ${message.id.slice(0, 8)} dispatch failed and session (${sessionId.slice(0, 8)}) existence could not be confirmed (opencode momentarily unreachable) \u2014 deferring this and later messages for conversation ${conv.id.slice(0, 8)} to the next tick rather than treating it as a genuine failure.`,
+            conversation_id: conv.id,
+            message_id: message.id
+          });
+          break;
+        }
+        const errorMessage = err instanceof Error ? err.message : String(err);
+        this.sessions.delete(conv.id);
+        this.supersede(conv.id, sessionId);
+        this.log({
+          level: "warn",
+          message: `Abandoning OpenCode session ${sessionId.slice(0, 8)} as the binding for conversation ${conv.id.slice(0, 8)} (it exists but failed to run a turn) \u2014 a fresh session is created on the next tick, whatever the persisted binding says by then.`,
+          conversation_id: conv.id,
+          message_id: message.id
+        });
+        await this.markFailed(conv.id, message.id, null, errorMessage).catch((markErr) => {
+          this.log({
+            level: "warn",
+            message: `markFailed PATCH for message ${message.id.slice(0, 8)} (conversation ${conv.id.slice(0, 8)}) failed (best-effort, not retried): ${markErr instanceof Error ? markErr.message : String(markErr)}`,
+            conversation_id: conv.id,
+            message_id: message.id
+          });
         });
         this.log({
           level: "error",
-          message: `Message ${message.id.slice(0, 8)} dispatch failed: ${err instanceof Error ? err.message : String(err)}`,
+          message: `Message ${message.id.slice(0, 8)} dispatch failed: ${errorMessage}`,
           conversation_id: conv.id,
           message_id: message.id
         });
-        continue;
+        break;
       }
       if (opencodeMessageId === null) {
         this.log({
@@ -2461,8 +2541,42 @@ var ChannelDriver = class _ChannelDriver {
     this.ensureWatcherRunning(sessionId);
     return dispatched;
   }
+  /**
+   * Record that `sessionId` is no longer a valid binding for `conversationId`
+   * (#553). Keyed by conversation and hard-capped, so it cannot grow with the
+   * number of failures — see the `supersededSessions` field doc.
+   */
+  supersede(conversationId, sessionId) {
+    this.supersededSessions.delete(conversationId);
+    this.supersededSessions.set(conversationId, sessionId);
+    while (this.supersededSessions.size > MAX_SUPERSEDED_CONVERSATIONS) {
+      const oldest = this.supersededSessions.keys().next().value;
+      if (oldest === void 0) return;
+      this.supersededSessions.delete(oldest);
+    }
+  }
+  /** Whether `sessionId` is the session this conversation has abandoned (#553). */
+  isSuperseded(conversationId, sessionId) {
+    return this.supersededSessions.get(conversationId) === sessionId;
+  }
+  /**
+   * Resolve the opencode session to run this conversation's turns in.
+   *
+   * `refusedSessionId` is set when the #553 guard fired — i.e. the persisted
+   * binding was an id this runner had abandoned, so a resurrection genuinely
+   * happened and a fresh session was bound instead. The caller reports it.
+   */
   async ensureSession(conv) {
     const bound = this.sessions.get(conv.id) ?? conv.opencode_session_id ?? null;
+    if (bound && this.isSuperseded(conv.id, bound)) {
+      this.log({
+        level: "warn",
+        message: `OpenCode session ${bound.slice(0, 8)} was abandoned for conversation ${conv.id.slice(0, 8)} after a failed dispatch but is still bound to it (the persisted id was written back by a turn already in flight) \u2014 ignoring it and binding a fresh session.`,
+        conversation_id: conv.id
+      });
+      this.sessions.delete(conv.id);
+      return { sessionId: await this.createAndBindSession(conv.id), refusedSessionId: bound };
+    }
     if (bound) {
       const exists = await sessionExists(this.port, bound);
       if (exists === false) {
@@ -2472,12 +2586,12 @@ var ChannelDriver = class _ChannelDriver {
           conversation_id: conv.id
         });
         this.sessions.delete(conv.id);
-        return this.createAndBindSession(conv.id);
+        return { sessionId: await this.createAndBindSession(conv.id) };
       }
       this.sessions.set(conv.id, bound);
-      return bound;
+      return { sessionId: bound };
     }
-    return this.createAndBindSession(conv.id);
+    return { sessionId: await this.createAndBindSession(conv.id) };
   }
   /**
    * Create a fresh OpenCode session for a conversation, cache the binding, and
@@ -2565,7 +2679,11 @@ var ChannelDriver = class _ChannelDriver {
    * (not-owned / out-of-range / deleted-at-source / workspace gone) / 413
    * (over-cap). On ANY non-2xx or thrown failure we return `null` so the caller
    * OMITS that one image and the text turn still sends — NEVER throws the turn.
-   * Failures are logged with context (no silent swallow).
+   * A 404 body carrying `{ reason: 'needs_reauth' }` (#547 — the server CONFIRMED
+   * a Slack `files:read` scope problem via `files.info`) instead resolves the
+   * `AttachmentFetchNeedsReauth` sentinel, so the in-thread note can steer the
+   * user to reconnect Slack instead of a generic "unavailable". Failures are
+   * logged with context (no silent swallow).
    */
   async fetchAttachmentDataUrl(messageId, index, mime) {
     try {
@@ -2574,6 +2692,25 @@ var ChannelDriver = class _ChannelDriver {
         { headers: { Authorization: this.getAuthHeader() } }
       );
       if (!res.ok) {
+        let reason;
+        try {
+          const body = await res.json();
+          if (body && typeof body.reason === "string") reason = body.reason;
+        } catch (parseErr) {
+          this.log({
+            level: "debug",
+            message: `Attachment fetch for message ${messageId.slice(0, 8)} index ${index}: error body was not JSON (${parseErr instanceof Error ? parseErr.message : String(parseErr)}) \u2014 treating as a plain failure`,
+            message_id: messageId
+          });
+        }
+        if (reason === "needs_reauth") {
+          this.log({
+            level: "error",
+            message: `Attachment fetch for message ${messageId.slice(0, 8)} index ${index} returned HTTP ${res.status} \u2014 server confirmed a Slack reauth/scope problem \u2014 omitting this image (text turn proceeds)`,
+            message_id: messageId
+          });
+          return { needsReauth: true };
+        }
         this.log({
           level: "error",
           message: `Attachment fetch for message ${messageId.slice(0, 8)} index ${index} returned HTTP ${res.status} \u2014 omitting this image (text turn proceeds)`,
@@ -2613,6 +2750,9 @@ var ChannelDriver = class _ChannelDriver {
     if (this.attachmentsSkippedSignalled.has(messageId)) return;
     this.attachmentsSkippedSignalled.add(messageId);
     const skippedReason = capabilityUnknown ? "unknown" : "unsupported";
+    const failedReason = outcomes.some(
+      (o) => o.status === "failed" && o.reason === "needs_reauth"
+    ) ? "needs_reauth" : void 0;
     this.log({
       level: "info",
       message: `Message ${messageId.slice(0, 8)}: ${skipped} image(s) skipped (${capabilityUnknown ? "capability was unreadable \u2014 failed open to text-only" : "model not attachment-capable"}), ${failed} image(s) unavailable (deleted-at-source or fetch failure) \u2014 noting to Evident`,
@@ -2622,7 +2762,8 @@ var ChannelDriver = class _ChannelDriver {
     void this.postSignal(conversationId, messageId, "attachments_skipped", {
       skipped,
       failed,
-      ...skipped > 0 ? { skipped_reason: skippedReason } : {}
+      ...skipped > 0 ? { skipped_reason: skippedReason } : {},
+      ...failedReason ? { failed_reason: failedReason } : {}
     });
   }
   /** Register a freshly-dispatched message with its session's watcher state. */
@@ -3909,6 +4050,32 @@ var ChannelDriver = class _ChannelDriver {
     }
     return messages;
   }
+  /**
+   * The `opencode_session_id` fragment of a status PATCH body — `{}` when this
+   * conversation has ABANDONED that session (#553). The field is optional
+   * server-side and an absent one leaves the persisted binding untouched, so
+   * omitting it is how a routine status write stops resurrecting it.
+   *
+   * ONLY for writes whose sole cost is a lost deep link. The `processing` notice
+   * degrades to no "View in Evident" link (the reaction swap still fires) and the
+   * turn-failure notice is built from the PATCH's own `error` text with a link off
+   * the persisted row — neither loses content the user came for. `markDone`
+   * deliberately does NOT use this helper: the server fetches the reply text
+   * THROUGH the session id it is given, so suppressing there would replace the
+   * agent's answer with a bare "✅ Done!" (the #183/#187 failure). The
+   * `ensureSession` guard, not this suppression, is what makes the self-heal
+   * stick.
+   */
+  sessionIdBody(sessionId, conversationId, messageId, status) {
+    if (!this.isSuperseded(conversationId, sessionId)) return { opencode_session_id: sessionId };
+    this.log({
+      level: "debug",
+      message: `Omitting the abandoned OpenCode session ${sessionId.slice(0, 8)} from the '${status}' update for message ${messageId.slice(0, 8)} so it is not re-bound to conversation ${conversationId.slice(0, 8)}`,
+      conversation_id: conversationId,
+      message_id: messageId
+    });
+    return {};
+  }
   /**
    * EXISTING combinedAuth route — now fired by the watcher on queued→running
    * (Task 3.3), NOT at dispatch/claim time. `{status:'processing',
@@ -3938,7 +4105,7 @@ var ChannelDriver = class _ChannelDriver {
         headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
         body: JSON.stringify({
           status: "processing",
-          opencode_session_id: sessionId,
+          ...this.sessionIdBody(sessionId, conversationId, messageId, "processing"),
           ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
           ...title ? { title } : {}
         })
@@ -3987,6 +4154,11 @@ var ChannelDriver = class _ChannelDriver {
         headers: { Authorization: this.getAuthHeader(), "Content-Type": "application/json" },
         body: JSON.stringify({
           status: "done",
+          // ALWAYS sent, even for a session this conversation has abandoned
+          // (#553): the server reads the reply text back out of THIS session id
+          // to deliver it. Omitting it would leave the user with "✅ Done!"
+          // instead of the answer — a worse regression than the resurrection it
+          // would prevent, which `ensureSession`'s guard handles anyway.
           opencode_session_id: sessionId,
           ...opencodeMessageId ? { opencode_message_id: opencodeMessageId } : {},
           ...title ? { title } : {},
@@ -4003,14 +4175,23 @@ var ChannelDriver = class _ChannelDriver {
   }
   /**
    * Mark a message `failed`. `sessionId` / `error` are threaded to the API ONLY
-   * when provided (issue #182): a bare `markFailed(conv, msg)` sends
-   * `{status:'failed'}` unchanged (the dispatch-failure path), while an errored
-   * OpenCode turn sends `{status:'failed', opencode_session_id, error}` so the
-   * failure reason reaches the channel.
+   * when provided (issue #182). Three states for `sessionId`:
+   *   - omitted (`undefined`)  → don't send the field, leave the persisted
+   *     session untouched (unused today; kept for API symmetry).
+   *   - a real id (`string`)   → send it, update the persisted session (the
+   *     turn-failure call sites: an errored OpenCode turn).
+   *   - explicit `null`        → send it, CLEAR the persisted session (issue
+   *     #485's dispatch-handoff-failure call site: the session id still
+   *     exists but is wedged, so the next attempt must get a fresh one
+   *     instead of reusing it — see WI-1's server-side null-clearing PATCH).
    */
   async markFailed(conversationId, messageId, sessionId, error2, usage) {
     const body = { status: "failed" };
-    if (sessionId !== void 0) body.opencode_session_id = sessionId;
+    if (sessionId === null) {
+      body.opencode_session_id = null;
+    } else if (sessionId !== void 0) {
+      Object.assign(body, this.sessionIdBody(sessionId, conversationId, messageId, "failed"));
+    }
     if (error2 !== void 0) body.error = error2;
     if (usage) Object.assign(body, usage);
     await this.callWithRetry(