The agentic rag, wrung out.
The rag for
Peter Steinberger’s OpenClaw Android app has been in Google Play review for more than a week. He asked for a Google contact (https://x.com/steipete/status/2106446147791597774). Sundar Pichai answered “Ack, will follow up,” and Android Ecosystem president Sameer Samat said he would get back quickly. That thread is the day’s story: 770k views on the ask, about 2 million on Sundar’s reply, and no ship yet. Runner-up for the product column is v2026.9.8, with the correction attached: GPT-6.1 Sol was announced and did not fully ship; 2026.9.9 is the promised patch.
Peter Steinberger asked publicly for a Google contact after OpenClaw’s Android app sat in Play review for more than a week. Sundar Pichai replied “Ack, will follow up.” Sameer Samat, President of the Android Ecosystem, followed with “Sorry for that. Will get back to you quickly.” No ship confirmation by end of day. The app is still in review.
Tony Simons published a one-page cheat sheet, docs checked 2026-10-03. SOUL.md is personality and tone. USER.md is the user’s profile and preferences. MEMORY.md is facts the agent has learned. AGENTS.md / .hermes.md is project instructions. SKILL.md is a reusable procedure. Session history is prior chats, not a notes file. Memory saves persist; the prompt snapshot refreshes in a new session. Passwords and API keys stay out of the notes.
Teknium posted a “Discord Setup, Checked For You” card: a wrong token is re-asked and never saved; Message Content Intent off gets a direct link to the toggle, then a re-check; the invite link prints with the right permissions; the owner is allowlisted without Developer Mode; a Discord section landed in the .env template. Same evening, Nous posted Linus Sebastian talking through a Hermes setup on the latest podcast (not sponsored, ~2 min clip). A reply in that thread said the Discord piece took three people a day. Nous answered that the Discord team has not taken their setup-stability suggestions, and that they had reached out so it should not take three days.
Official v2026.9.8 (43 PRs, 8 contributors named on the post; docs say the same scale). Delegated agent results return to the conversation that asked, once: inline if settled, or a single later delivery. Internal sessions no longer finish on NO_REPLY. Codex catalog processes hold fewer in-memory shell keys. Windows updates retry file locks; macOS npm updates accept aliased install paths. Docs are explicit that GPT-6.1 Sol was prepared (1.05M context, 128k output, 272k active input budget in OpenClaw) and was not complete in this release. Hannes Rudolph confirmed the miss: the bundled Codex CLI is 0.158.0 and needs a newer build. He posted three workarounds (updated ChatGPT app for computer use, force the OpenClaw runtime, or pin a 0.159.1+ Codex binary) and said 2026.9.9 is the fix. Not shipped by end of day.
Maurice (KI-Vater) posted a before/after of Hermes Plugin Guard v0.2.1 on a synthetic plugin. Removing `verify=False` clears the TLS warning. Outbound-call hints stay on purpose: a network call needs review and is not auto-malicious. The scan does not import or run the target. He is explicit that a pass is not a safety guarantee, and asked for real rule-id examples of useful vs noisy warnings.
Luke’s Oct 3 digest covers 175 merges from Oct 2. Hermes Release Watch pulled the operator set: catalog adds include Gmail, WhatsApp, Apple Health, Agent Builder, and several memory providers. Supermemory and Honcho install from the catalog; core no longer bundles Honcho. Memory-provider Python deps migrate on Desktop, the gateway, and scripted updates without a terminal. Plugin-guard denylist noise no longer blocks or prompts. Local MCP and plugin tools in one tool_call run instead of being rejected. SMS, email, and WeCom deliver long replies and cron output in full. Desktop skips rebuilds when HEAD did not move. Windows tray restore no longer leaves the window dead to input.
Never miss a night
One feed for everything, or one per agent harness you run. Rebuilt every night, with full credit on every item.
