CEREBRO
machine-read, human-curated
Group signals, write briefing now.
Agentic SaaS: Claude Tag
Anthropic ships Claude Tag, tag Claude into Slack channel, joins as teammate w/ own identity+memory, proactive not reactive bot. Under hood: Claude Code engine, each thread gets own sandboxed instance (clone/write/test/compile), thrown away after (spins up isolated sandbox per thread). Internal dogfood: 65% of Anthropic product team's new code now written via internal Claude Tag (bcherny). Use cases surfacing: PR writing, incident investigation, data analysis, org search-engine replacement, answers not links (bcherny thread); proactive monitoring + emoji-react thread resolution (bcherny). Beta today, Slack only, Claude Enterprise/Team, more surfaces coming (bcherny).
Related: Orca, ADE for running fleet of parallel agents (Codex/Claude Code/OpenCode/Pi) each own worktree, one dashboard, own subscription, usage tracking across accounts.
Coding agents / harness internals
Claude Code sends 33k tokens before reading your prompt vs OpenCode's 7k, measured system-prompt+tool-schema+scaffolding overhead on identical model/machine/task, community read: Anthropic optimizes capability over token cost. Direct signal for your caching/context-budget tracking.
Claude Code v2.1.204 released.
system_prompts_leaks repo updated: extracted system prompts for Fable 5, Opus 4.8, Claude Code, GPT-5.6/Codex, Gemini 3.5, Grok, Cursor, Copilot, more, regularly refreshed reference for prompt-engineering comparisons.
destructive_command_guard, hook blocking dangerous git/shell commands before execution across Claude Code, Codex CLI, Gemini CLI, Copilot CLI, Cursor, worth eval for your own guardrails.
LLM mechanics / research
GLM-5.2 step change for open agents, capability threshold watch on open-weight agentic models, paired w/ new paper on open RL recipes for terminal agents.
Lilian Weng's 35-paper roundup on Harness Engineering for RSI, condensed survey of recursive self-improvement / agent-harness research direction.
Frontier post-training recipe review w/ Finbarr Timbers, state-of-play on RLHF/post-training recipes heading toward frontier-scale from Olmo-style baseline.
Mechanistic interpretability applying causality theory to LLMs, community pushback: can't yet distinguish real reasoning from input-output mapping, treat "understanding" claims skeptically.
Fable 5 field guide, community stress-testing model limits before subscription subsidy ends.
Some ideas for what comes next, May 2026, Gemini Flash 3.5/Mythos/open-closed balance/open-source surge landscape overview.
Vibe-coding & agent repos
background-agents (Open-Inspect), open-source hosted background coding agent, full dev env (Node/Python/git/browser automation/VS Code), reachable via web/Slack/GitHub PRs/Linear/webhooks, multiplayer sessions.
page-agent, in-page JS GUI agent, no extension/headless browser needed, text-based DOM manipulation not screenshots, control any webpage w/ natural language.
awesome-llm-apps, 100+ runnable AI agent/RAG app templates across Claude/Gemini/OpenAI/xAI/Qwen/Llama, quick-start clone-and-ship reference.
Signals
- We're launching Claude Tag today. Tag Claude into Slack and it works in channel with you. It’s proactive, multiplayer, with its own identit…
- Introducing Claude Tag, a new way for teams to work with Claude. In Slack, Claude joins as a team member with access to the channels and to…
- This is the start of Claude Everywhere. It’s Claude Code under the hood so it’s just as good at writing code. 65% of our product team’s new…
- I also have Claude monitoring Slack channels to proactively respond to people to answer their questions and draft PRs, and to react with sp…
- I tag Claude many times a day to write PRs, address user feedback, investigate incidents, do data analyses, summarize information, etc. Use…
- ColeMurray/background-agents: An open-source background agents coding system
- stablyai/orca: Orca is the ADE for working with a fleet of parallel agents. Run any coding agent with your own subscription. Available on d…
- (AINews) The Field Guide to Fable
- v2.1.204
- Claude Code sends 33k tokens before reading the prompt; OpenCode sends 7k
- Tag Claude in a channel, it spins up an instance with its own sandbox. It clones repos, writes code, tests, compiles all in that isolated e…
- Dicklesworthstone/destructive_command_guard: The Destructive Command Guard (dcg) is for blocking dangerous git and shell commands from bein…
- Shubhamsaboo/awesome-llm-apps: 100+ AI Agent & RAG apps you can actually run — clone, customize, ship.
- That's it. Point it at a channel, give it a task, and let it work. It's in beta on Slack today for Claude Enterprise and Team customers. Mo…
- https://cm-472c2900564d4cb7-site-e06d68085bab.bubbleupdappos.workers.dev/
- http://OKX.ai
- Claude + MCP Hookami
- GLM-5.2 is the step change for open agents
- asgeirtj/system_prompts_leaks: Extracted system prompts from Anthropic - Claude Fable 5, Opus 4.8, Claude Code, Claude Design. OpenAI - Cha…
- Claude is really proactive with Claude Tag. You don’t need to prompt it to do work, it can do work proactively based on your instructions.…
- alibaba/page-agent: JavaScript in-page GUI agent. Control web interfaces with natural language.
- Mechanistic interpretability researchers applying causality theory to LLMs
- (AINews) Lilian Weng summarizes 35 papers on Harness Engineering for RSI
- Frontier post-training recipe review with Finbarr Timbers
- Some ideas for what comes next, May 2026
Sources & citations
| Source | Fetched | In briefing |
|---|---|---|
| x | 186 | 10 |
| rss | 140 | 6 |
| github | 47 | 6 |
| hackernews | 60 | 2 |
| 25 | 1 | |
| show_hn | 40 | 0 |
| yc-rfs | 16 | 0 |
| yc-launch | 3 | 0 |
| gmail | 0 | 0 |
| total | 517 | 25 |