@holyglory/codex 0.154.0-multi.1 → 0.156.1-multi.1

package/README.md

@@ -161,6 +161,19 @@ codex account auto off
 codex account priority set-all 1000
 ```
 
+Apply one banked reset to a specified profile with `codex account reset work`.
+The command applies immediately and selects the available credit that expires soonest;
+non-expiring credits are used last. Use `--credit-id CREDIT` to choose one explicitly,
+or `--json` for the outcome, windows reset, request identity and refreshed usage.
+
+If the response is uncertain, retry the same operation with
+`codex account reset work --credit-id CREDIT --request-id REQUEST`, using the
+identifiers printed by the original attempt. The replay reaches the service even
+when that credit has already been consumed; it does not select another credit.
+Confirmed resets and explicit no-op outcomes return zero. No credit, unsupported
+authentication, request failures and uncertain outcomes return nonzero. A refresh
+failure is reported separately and does not undo a confirmed reset.
+
 ### Inspect and maintain accounts
 
 | Command                                         | Purpose                                                           |
@@ -209,6 +222,41 @@ Enter these in the interactive Codex prompt, not your shell:
 
 You can also ask the agent to list local account aliases, identify which profile handled the current turn, show available limits, or change priorities. The built-in `account_management` tool supports these bounded operations; it does not sign in, activate, rename, or remove profiles and does not reveal credentials, email, or service/workspace identifiers. Use the explicit account commands for those management actions.
 
+## Diagnose network failures
+
+Codex Multi keeps model HTTP and WebSocket diagnostics in a separate private
+`network_diagnostics_1.sqlite` database beside its other SQLite state files
+(normally in `~/.codex`). Records have no automatic expiry or per-task row cap;
+ordinary debug-log pruning and runtime restarts do not remove them.
+
+```sh
+codex debug network-errors --thread-id TASK_ID --limit 20
+codex debug network-errors --thread-id TASK_ID --before-id NEXT_BEFORE_ID
+codex debug network-errors --thread-id TASK_ID --include-context
+```
+
+The JSON report includes UTC Unix timestamps in milliseconds, task and turn IDs,
+available request/response IDs, HTTP statuses, provider error codes, WebSocket
+close codes and redacted reasons, retry schedules, fallback, and completion
+markers. `--since-ms` limits the time range; `nextBeforeId` continues an older page.
+The default view shows incidents; `--include-context` adds successful request and
+completion records around them. Incident queries use dedicated indexes, so a long
+successful run does not bury earlier failures or require scanning all its activity.
+
+Large WebSocket requests upload complete context items in ordered batches before
+generating one response. The client targets 4 MiB per request; this is a batching
+choice, not the server's documented limit. If an item cannot fit intact or the
+server rejects preparation, Codex uses HTTP with the complete request. No images
+or conversation items are discarded. The ledger records request byte counts and
+batch preparation alongside recovery events.
+
+Request/response bodies, prompts, tool output, credentials, and arbitrary headers
+are excluded. Individual error/reason fields retain up to 8,192 characters, with
+an explicit truncation marker; metadata fields retain up to 512. Disk write
+failures are reported and retried without stopping model work. Queue overflow is
+reported immediately and counted in the next retained record. The ledger grows
+with network activity and is not uploaded automatically.
+
 ## Inspect local usage
 
 Local accounting answers **what this fork observed on this machine**, not how much quota remains on OpenAI's service. Use `codex account limits` for service limits. The local collector does not reconstruct a complete usage history from before it was installed.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@holyglory/codex",
-  "version": "0.154.0-multi.1",
+  "version": "0.156.1-multi.1",
   "description": "Unofficial multi-account fork of the OpenAI Codex CLI.",
   "license": "Apache-2.0",
   "bin": {
@@ -29,11 +29,11 @@
   },
   "packageManager": "pnpm@10.34.5+sha512.a4ee05f2f73658255bd6a89859c065a45c28a57daefae2c893a168ee2b73168c37b91e83e57ea67654ad03f03031746430e8bce38e362e042605fb8abc80192e",
   "optionalDependencies": {
-    "@holyglory/codex-linux-x64": "npm:@holyglory/codex@0.154.0-multi.1-linux-x64",
-    "@holyglory/codex-linux-arm64": "npm:@holyglory/codex@0.154.0-multi.1-linux-arm64",
-    "@holyglory/codex-darwin-x64": "npm:@holyglory/codex@0.154.0-multi.1-darwin-x64",
-    "@holyglory/codex-darwin-arm64": "npm:@holyglory/codex@0.154.0-multi.1-darwin-arm64",
-    "@holyglory/codex-win32-x64": "npm:@holyglory/codex@0.154.0-multi.1-win32-x64",
-    "@holyglory/codex-win32-arm64": "npm:@holyglory/codex@0.154.0-multi.1-win32-arm64"
+    "@holyglory/codex-linux-x64": "npm:@holyglory/codex@0.156.1-multi.1-linux-x64",
+    "@holyglory/codex-linux-arm64": "npm:@holyglory/codex@0.156.1-multi.1-linux-arm64",
+    "@holyglory/codex-darwin-x64": "npm:@holyglory/codex@0.156.1-multi.1-darwin-x64",
+    "@holyglory/codex-darwin-arm64": "npm:@holyglory/codex@0.156.1-multi.1-darwin-arm64",
+    "@holyglory/codex-win32-x64": "npm:@holyglory/codex@0.156.1-multi.1-win32-x64",
+    "@holyglory/codex-win32-arm64": "npm:@holyglory/codex@0.156.1-multi.1-win32-arm64"
   }
 }