Glean 拾遗
中 EN
Submit a link
Daily · timeline

A few picks a day.

Mon, Jun 8, 2026 4picks
← 06-07
Calendar ▾
2026 · 06
MoTuWeThFrSaSu 123456789101112131415161718192021222324252627282930
has picks today
06-09 →
10:57

Maple: An Open-Source Observability Platform Built on OpenTelemetry and ClickHouse

Maple：基于 OpenTelemetry + ClickHouse 的开源可观测性平台

Maple is an open-source observability platform for traces, logs, and metrics, built on OpenTelemetry and ClickHouse. It features an OTLP ingest gateway with key-based auth, a chat agent, alert evaluation, and SQLite/Turso-backed dashboard persistence. The monorepo ships with Clerk or self-hosted auth modes and a full suite of CI/CD workflows for Cloudflare Workers, targeting teams that want to own their observability stack in a TypeScript-native codebase.

github.com · 9 min · ClickHouse · Cloudflare · Monorepo · Observability · OpenTelemetry · TypeScript
06:00

10 Lessons for Writing a Good AGENTS.md for Codex and Claude Code

AGENTS.md 写作十诫：让 AI 编码代理真正听懂你的项目

Ten hard-won lessons from running Codex and Claude Code side by side, distilled into a survival guide for writing AGENTS.md files that actually work. Key moves include capping the root file at 200 lines, listing what not to introduce alongside the actual stack, writing rules the tool can mechanically check instead of vague principles like “keep it simple”, and treating the entry file as a router to architecture docs rather than a single dump. Other high-leverage practices involve using PLANS.md to break long-running tasks into reversible phases inside an isolated worktree, giving high-risk directories their own local guardrails, layering intent–intercept–permission–sandbox so red lines aren't left to model memory alone, storing auditable long-term memory in MEMORY.md with a 30‑day hurdle, and separating personal style from team conventions from machine permissions. The guide closes with a copy‑ready skeleton and the principle that the entry file should grow like a test suite every time the tool gets something wrong.

x.com · 24 min · Agents · AI
06:00

The Orchestration Tax: Why 20 AI Agents Don't Mean 20x Output

AI 代理的编排税：为什么开 20 个 agent 不等于 20 倍产出

Addy Osmani coins 'orchestration tax': spawning agents is cheap, but closing the loop—reviewing, judging, merging—is serial and bounded by your cognitive bandwidth. Using Amdahl's Law and Python's GIL as analogies, he argues you are the single-threaded bottleneck in an otherwise concurrent system. Tactics: cap parallelism to your review rate, split work into delegable vs. judgment-heavy piles, batch reviews to cut context-switch costs, and force agents to self-verify. Aimed at engineers who run multiple AI agents daily and feel busy but unproductive.

x.com · 9 min · Agents · Performance
06:00

Claude API adds auto-caching: single cache_control param cuts input cost to 10%

Claude API 新增自动缓存：用 cache_control 参数一行切到 1/10 成本

Anthropic introduced prompt auto-caching in the Claude Messages API. Instead of manually moving breakpoints across conversation turns, a single top-level `cache_control: {type: 'ephemeral'}` auto-places the cache at the last cacheable block. Cached tokens cost 10% of base input price and reduce prefill latency. Ideal for agents and coding assistants where most context remains identical turn-over-turn. The post cites Manus founder @peakji on cache hit rate being the most critical metric for production agents, and links to Claude Code's cache-friendly prompt design insights.

x.com · 5 min · Agents · LLM