Topic hub
Context engineering
How teams package memory, rules, traces, tools, and task state so agents can work without restarting from raw chat context.
Retrieval answer
How teams package memory, rules, traces, tools, and task state so agents can work without restarting from raw chat context. Context engineering is the work of turning scattered organizational knowledge into durable inputs an agent can inspect and act on. The strongest systems separate reusable rules, task state, source evidence, and operator corrections instead of burying everything in one prompt.
Pattern memory
What patterns are emerging?
- medium
Skills become a portable capability layer
Agent skills are emerging as a portable capability layer, but their value depends on progressive disclosure, provenance, security review, and behavioral evaluation.
Field notes
What should readers understand next?
A Vector Store Is Not An Agent Memory System
Contextual AI separates working, procedural, semantic, and behavioral memory, with evaluation and provenance gates protecting every durable write.
ChatGPT Cuts Repeated Work Across The Agent Stack
ByteByteGo's OpenAI engineering walkthrough connects persistent sessions, stable prompt prefixes, deferred tools, delta tokenization, cache-aware routing, and split inference.
Anthropic Moves Large Tool Libraries Out Of Context
Anthropic's Tool Search Tool, Programmatic Tool Calling, and Tool Use Examples separate discovery, orchestration, and usage guidance for agents with large tool libraries.
Dr. Skill Audits What An Agent Loads Before It Works
Dr. Skill scans skills and MCP servers for collisions, duplication, secrets, drift, missing metadata, and unused loadout, with local and CI-friendly commands.
Genkit Adds Progressive Disclosure For Agent Skills
Genkit now loads Agent Skills through middleware that discovers SKILL.md metadata first and activates full instructions, references, and scripts only when needed.
Parallel Packages Web Research As A Responses-Compatible Subagent
Parallel's Responses API offers cited web research behind an OpenAI-compatible endpoint, with bounded effort tiers, streaming, and stateful follow-ups.
Firecrawl MCP Turns Web Search into a Bounded Agent Capability
Firecrawl's MCP launch points to a cleaner web-context surface for agents: OAuth for humans, API-key headers for server jobs, and keyless trials for low-friction testing.
Mem0 Shows Agent Memory Should Not All Live in Context
Mem0's Claude Code experiment separates durable memory from the conversation window: retrieve the relevant slice, survive /clear, and avoid loading every memory file up front.
OpenAI Shows Efficiency Is a Full-Stack Agent Problem
OpenAI's GPT-5.6 efficiency write-up connects model training, inference optimization, and the Codex/ChatGPT Work harness into one compounding cost-performance loop.
OpenAI's ARC-AGI-3 Jump Was a Harness Result
OpenAI's ARC-AGI-3 write-up shows why agent benchmarks measure the model plus the runtime harness: retained reasoning and compaction changed both score and token use.
LangChain's Data Agent Turns BI Into a Context Maintenance Loop
LangChain's agent-first data stack shows that reliable data agents depend on maintained context layers, trust signals, observability, and data-team feedback loops.
Audit Claude Code Context Before You Rewrite the Prompt
The safest response to Claude Code prompt bloat is an evidence-led context audit: inspect loaded memory, skills, hooks, MCP tools, and setting precedence before deleting safeguards.
AI-Native Engineering Teams Are Smaller and Stricter
AI-native engineering changes team size, roles, onboarding, metrics, and review: code is no longer written only by humans, but context and quality ownership get stricter.
Claude Code Subagents Split Memory Instead of Inflating One Session
Claude Code subagents show a practical form of agent memory hygiene: move noisy work into a separate context and return only the compressed result to the main session.
Unabyss Builds a Shared Context Layer Across Claude, Codex, and Cursor
Unabyss points to where agent tooling is going: context stops being one client's internal memory and becomes a separate product layer with governance and portability.
Raw signals
What changed recently?
Firecrawl MCP Cuts Context Use for Web Tools
Firecrawl's MCP docs expose web search, scrape, parse, and interact work through bounded agent access paths for OAuth users, servers, and keyless trials.
Google Gemini: Create, edit, and summarize with Gemini on macOS using your voice.
A public X post from Google Gemini with a linked primary source flags Create, edit, and summarize with Gemini on macOS using your voice. Stay in your flow by speaking into any active window: Dictate clean text or ask Gemini to transform highlighted...
mem0: Experiment 2: Does a preference survive a hard context wipe?
A public X post from mem0 with a linked primary source flags Experiment 2: Does a preference survive a hard context wipe? We told Claude(mid-conversation) to write functions with standard for loops instead of list comprehensions. Then ran /...
OpenAI: We’re also upgrading Auto-review in the ChatGPT app and Codex CLI from GPT-5.4 to GPT-5.6 Luna.
A public X post from OpenAI with a linked primary source flags We’re also upgrading Auto-review in the ChatGPT app and Codex CLI from GPT-5.4 to GPT-5.6 Luna. Combined with Luna’s new price, we expect Auto-review to cost about 10x less, makin...
Google Gemini: 1) Contextual Understanding With Gemini Omni’s real world understanding, Omni can understand i...
A public X post from Google Gemini with a linked primary source flags 1) Contextual Understanding With Gemini Omni’s real world understanding, Omni can understand instructions inside your reference video. For example, you can show written instructio...
LangChain: OpenWiki now connects directly to LangSmith tracing projects to provide better context into ho...
A public X post from LangChain as a public source in its own right flags OpenWiki now connects directly to LangSmith tracing projects to provide better context into how coding agents are actually interacting with your codebase. To generate the most use...
OpenAI: A benchmark score reflects the model as well as the harness and settings used to run it.
A public X post from OpenAI with a linked primary source flags A benchmark score reflects the model as well as the harness and settings used to run it. For long-running agents, retaining reasoning and compacting context lets the model build o...
LangChain: Our data agent now handles roughly 40x the request volume our 3 person data team could manage...
A public X post from LangChain with a linked primary source flags Our data agent now handles roughly 40x the request volume our 3 person data team could manage directly. Now, our data team can focus on the models, context, and guardrails that ma...
OpenAI Developers: We're introducing two new transcription models in the API: • GPT-Live-Transcribe: built for lo...
A public X post from OpenAI Developers with a linked primary source flags We're introducing two new transcription models in the API: • GPT-Live-Transcribe: built for low-latency live transcription. • GPT-Transcribe: optimized for asynchronous transcript...
Firecrawl: Firecrawl is now an official @Replit Connector 🔥 Get high quality web context into your Replit...
A public X post from Firecrawl with a linked primary source flags Firecrawl is now an official @Replit Connector 🔥 Get high quality web context into your Replit apps with Firecrawl /search, /scrape, and /interact, all set up in a single click. C...
OpenAI: Health in ChatGPT is starting to roll out to U.S.
A public X post from OpenAI with a linked primary source flags Health in ChatGPT is starting to roll out to U.S. users. You can securely connect Apple Health and supported medical records to understand your information in context, track what...
Firecrawl: Introducing our most accurate /search yet.
A public X post from Firecrawl with a linked primary source flags Introducing our most accurate /search yet. We trained a model to return the excerpts that best answer your query, giving agents highly relevant context from each result. It's SOTA...
LangChain: We’re launching the Eval Engineering Skill, a skill that helps coding agents build evals using...
A public X post from LangChain as a public source in its own right flags We’re launching the Eval Engineering Skill, a skill that helps coding agents build evals using context from a repository + agent traces. Everything you need to know from @vtrivedy...
OpenCode: Hy3 now available in Go text · 256K context latest model from Tencent
A public X post from OpenCode as a public source in its own right flags Hy3 now available in Go text · 256K context latest model from Tencent
Google Gemini: Gemini 3.6 Flash can quickly understand context and help you create your own custom creative t...
A public X post from Google Gemini with a linked primary source flags Gemini 3.6 Flash can quickly understand context and help you create your own custom creative tools. For example, it can extract real natural textures from photos of your environme...
OpenCode: Gemini 3.6 Flash and 3.5 Flash Lite now available in OpenCode - 1M context - 3.6 Flash: 17% ch...
A public X post from OpenCode as a public source in its own right flags Gemini 3.6 Flash and 3.5 Flash Lite now available in OpenCode - 1M context - 3.6 Flash: 17% cheaper output than 3.5 Flash - 3.5 Flash Lite: 80% cheaper than 3.5 Flash
OpenCode: Laguna S 2.1 is now free on OpenCode 1M Context · fully open source Poolside's most capable mo...
A public X post from OpenCode as a public source in its own right flags Laguna S 2.1 is now free on OpenCode 1M Context · fully open source Poolside's most capable model to date
Agent skills need behavioral evals, not prose review
Benchmarks show that expert-authored skills can help while self-generated skills can underperform a no-skill baseline.
Augment Code: For the first time ever: Grok 4.5 is now available.
A public X post from Augment Code as a public source in its own right flags For the first time ever: Grok 4.5 is now available. Welcome @SpaceXAI to Cosmos, our agent orchestration platform. Together with Augment's context engine, Grok 4.5 is a powerful o...
Retrieval and long context do not create company memory
Finding a relevant fragment does not establish whether it is current, authorized, trustworthy, or applicable to the present task.
Claude Code Startup Context Can Be Reduced to 13K Tokens
A measured context audit shows that tool descriptions and default instructions consume a material share of the initial window before repository work begins.
Codebase Memory Combines Search with a Repository Map
A code retrieval tool adds structural repository context to text search so an agent can locate relevant code without loading the whole project into context.
Pydantic AI loads capabilities only when needed
A capability can bundle instructions, tools, model settings, and hooks while exposing only a compact description until activation.
CLAUDE.md Shrinks as Skills, Hooks, and Subagents Mature
Agent behavior moves out of one giant context file into scoped mechanisms that load only for the relevant task and lifecycle stage.