CEREBRO
machine-read, human-curated
Researched and written with 252,187 tokens (in 8 · out 25,299 · cache-read 49,251 · cache-create 177,629) across 4 Claude calls
Coding agents: Claude Code and tooling
- v2.1.316: adds
CLAUDE_CODE_DISABLE_ULTRATHINKto turn off theultrathinkkeyword, plus pooled group budgets on the Claude apps gateway (pooled_budgeton the spend limits admin API) so a group shares one monthly cap. - v2.1.314: adds
permissions.blockWritesOutsideWorkingDirectories, which makes Write/Edit/NotebookEdit refuse out-of-tree paths in every permission mode, and an IdPgroupsfilter on gateway upstreams. - You can now build custom themes for Claude Code on desktop with mods! Themes can change colors and add animated backgrounds, with a
theme-examplesplugin to start from. - We're removing the 5-hour limit for verified Ultra accounts. Ultra users no longer hit the rolling 5-hour usage window.
- anthropics/skills: Anthropic's reference repo for Agent Skills (folders of instructions, scripts and resources loaded on demand), useful as the canonical pattern for your own skills.
- My Claude Code config (CLAUDE.md, hooks, 10 subagents) is public. Which parts would you cut? A public config teardown, worth reading the critiques for what to prune from your own setup.
- My MCP server just got approved for Claude's connector directory. A write-up of every blocker hit during directory approval and its fix, a useful checklist if you ship an MCP server.
- Opus 5.5 running for more than 24 hours for a refactoring task Anecdotal long-horizon run data point on autonomy and cost for large refactors.
Token, cache and context economics
- I measured how much each "save tokens" trick actually costs on a Pro plan. Measured result: returning to an old chat after 1h costs about 20x more, which is a cache-expiry effect (the prompt cache has lapsed, so the whole context is re-billed).
- What Claude Code plugins actually save tokens? Community thread comparing which plugins really cut usage, to read alongside the measurement above.
- JuliusBrussee/caveman: skill plus proxy that claims about 65% fewer tokens by making the agent talk like a caveman, and it now pitches itself as a full coding agent with benchmark claims against Codex and Claude Code (vendor-run, treat sceptically).
- CodeMesh: Product Hunt launch claiming 90% lower AI coding costs, with no detail given on the mechanism.
- A frontier model reviews, a cheaper one codes: my Claude Code workflow with Quarterdeck A split-model workflow: cheap model implements, frontier model reviews, which matches the cost-tiering approach.
LLM mechanics: sparse attention
- Tencent-Hunyuan/FlashTopK (paper: arXiv 2610.32057, announcement): a fused kernel for the DeepSeek-style sparse-attention indexer that never writes the score table to memory. It cuts Top-K at 1M context from 5.32s to 1.62s, stays exact and bit-deterministic, and supports GLM-5 and Hy4-preview.
Agent orchestration and workflows
- Tencent/LoopForge: self-hosted workspace that runs Claude Code, Codex and CodeBuddy in Docker sandboxes with project roles, scripted SOPs and human approval gates, ending in a PR.
- RT @Steve_Yegge: software factories talks Yegge's contrarian take: pace AI output to human speed rather than maximising throughput.
- RT @OpenAIDevs: Pietro Schirano's agents working together A multi-agent collaboration idea prototyped with Codex in a couple of hours.
- Simon Willison on a case study of agentic engineering Flagged as the sharpest-edge example of agentic engineering "in the Fable era", worth reading for the limits of what's possible.
Agent design essays (cryptologie.net)
- The deterministic/stochastic border: where deterministic code ends and agentic code begins, and how that line keeps moving.
- Periodic coding-agent queries for your codebases: running scheduled agent queries against repos as a standing audit mechanism.
- The nursing problem: infra agents and their slow crystallization: agents that live inside your infrastructure and how their behavior gradually hardens into fixed code.
Agentic SaaS and protocols
- ag-ui-protocol/ag-ui: open, event-based protocol standardizing how agents connect to user-facing apps, a candidate for the agent-to-frontend layer.
- Theo on T3 Code: a non-technical user asked her agent to enable a nightly-only feature, and the agent cloned the source so she is now on a fork, a sign of agents casually forking software to customize it.
Signals
- v2.1.316
- v2.1.314
- JuliusBrussee/caveman: 🪨 why use many token when few token do trick. Viral skill + proxy for coding agents that cuts 65% of tokens by talk…
- I measured how much each "save tokens" trick actually costs on a Pro plan. Coming back to an old chat after 1h costs ~20x more.
- Tencent/LoopForge: Self-hosted workflows for coding-agent delivery. Orchestrate CodeBuddy, Codex and Claude Code with scripts and human app…
- anthropics/skills: Public repository for Agent Skills
- My Claude Code config (CLAUDE.md, hooks, 10 subagents) is public. Which parts would you cut?
- https://cryptologie.net/posts/deterministic-stochastic-border/
- https://cryptologie.net/posts/periodic-coding-agent-queries-for-your-codebases/
- What Claude Code plugins actually save tokens?
- https://cryptologie.net/posts/the-nursing-problem-infra-agents-and-their-slow-crystallization/
- A frontier model reviews, a cheaper one codes: my Claude Code workflow with Quarterdeck
- RT @Steve_Yegge: I just finished talks about my software factories at two conferences. Afterwards a bunch of people came up and told me wha…
- RT @OpenAIDevs: Pietro Schirano (@skirano) dreamed up a way for agents to work together. The next morning he prototyped it with Codex in a…
- My MCP server just got approved for Claude's connector directory. Every wall I hit on the way, and the fix for each
- CodeMesh
- http://flashtopk.select
- I have a nontechnical friend who has been using T3 Code a lot lately. She saw this, got hyped, and immediately asked her agent to customize…
- Fascinating case study of the sharpest edge of what you can do with agentic engineering in the Fable era - my eyebrows raised at least a do…
- Opus 5.5 running for more than 24 hours for a refactoring task
- ag-ui-protocol/ag-ui: AG-UI: the Agent-User Interaction Protocol. Bring Agents into Frontend Applications.
- Tencent-Hunyuan/FlashTopK
- https://arxiv.org/abs/2610.32057
- We're removing the 5-hour limit for verified Ultra accounts.
- You can now build custom themes for Claude Code on desktop with mods! Change the colors, add an animated background, and make it your own.…
Sources & citations
| Source | Fetched | In briefing |
|---|---|---|
| x | 263 | 12 |
| 25 | 6 | |
| github | 36 | 4 |
| rss | 140 | 3 |
| show_hn | 40 | 0 |
| github_search | 20 | 0 |
| swipe | 20 | 0 |
| yc-rfs | 11 | 0 |
| yc-launch | 3 | 0 |
| gmail | 1 | 0 |
| seed_urls | 1 | 0 |
| crackscan | 0 | 0 |
| hackernews | 0 | 0 |
| ossinsight | 0 | 0 |
| reddit_users | 0 | 0 |
| total | 560 | 25 |