Lora
中文

Writing

Notes on systems I am building, checking, and learning from.

Browse 86 English articles and 93 Chinese articles. English translations link back to their original versions.
Filter articles All articles
86 articles across all sections
Featured note English
Learning guides

Use When Agents Slow Down, Reflexion, Hyperband, and agent evaluation to compare continued revision with independent attempts under reproducible budgets.

Read this note

A starting point

Selected writing

3 notes from the collection
  1. 01 What Tau solves by splitting an AI agent into three layers Tau separates model adapters, the agent core, and the application so tools and frontends can work together without depending on each other. Read this note
  2. 02 Connecting an agent loop to a frontend with events Follow tool execution and repeated model calls, separate streamed deltas from final messages, and explore frontend state with a model-free Python demo. Read this note
  3. 03 AI Engineering Harness: verifiable delivery for AI code A multi-agent engineering workflow uses isolated worktrees, CI, cold-start review, evidence gates, and layered context to control delivery of AI-generated code. Read this note

Ideas to revisit

Long reads and practical guides

69 articles

  1. EmDash 1.0 separates plugin catalogs from publishing

    EmDash 1.0 separates author identity, signed releases, plugin discovery, and runtime permissions. Installers verify records independently.

    Projects & practice Developer tools MCP & skills Security & privacy Web development
    English

  2. Agents API browser auth keeps passwords out of model input

    How Agents API separates site approval, login credentials, and screenshots, and why network access and site approval cannot replace action-level confirmation.

    Technical articles Agents Security & privacy Runtime & infrastructure
    English

  3. Instant agent steering and recoverable tool calls

    A close reading of Codex instant_interrupt: ending model output, continuing background tools, and preserving state when users change requirements mid-run.

    Learning guides Agent engineering Runtime & infrastructure Memory & context
    English

  4. Cloudflare redesigns its CLI for agents

    Cloudflare launches cf in open beta, using Wrangler usage data to explain default JSON, command search, and typed configuration.

    Technical articles Agents Developer tools MCP & skills
    English

  5. OpenShell enforces agent permissions outside the workload

    NVIDIA OpenShell places sandbox, network, credential, and policy controls outside the agent workload, with a Supervisor enforcing permission boundaries.

    Projects & practice Security & privacy Agent engineering Runtime & infrastructure
    English

  6. Keep review evidence separate from compacted agent context

    Independent Codex Guardian history shows why generation context and review evidence need separate records, with provenance through resume and rollback.

    Learning guides Agent engineering Memory & context Evaluation Security & privacy
    English

  7. Cloudflare data puts agent traffic on the website budget

    Cloudflare and Fastly network data reveal repeated crawling, origin costs, access-policy questions, and the returns websites get from machine traffic.

    Technical articles Agents Web development Runtime & infrastructure
    English

  8. MentraOS runs Miniapps on phones for multiple glasses

    MentraOS runs Miniapps in a phone runtime, sharing a connection to manage input, output, hardware capabilities, and permissions across smart glasses.

    Projects & practice Multimodal systems Runtime & infrastructure Product design
    English

  9. Why old CLAUDE.md rules can hinder a model upgrade

    Use prompt audits, instruction layers, and migration evals to retain real requirements, reassess model workarounds, and enforce hard limits in code.

    Learning guides Agent engineering Evaluation Developer tools
    English

  10. Company Brain separates team memory permissions

    Company Brain scopes team-agent memory by room and user permissions, with personal authorization, traceable sources, and separate tool-write access.

    Projects & practice Memory & context Security & privacy Agents
    English

  11. Microsoft connects AI-built tools to app delivery

    After Copilot Code builds an internal tool, Managed Runtime handles Git, dev environments, Entra ID, connectors, previews, deployment, and governance.

    Technical articles Developer tools Web development Runtime & infrastructure
    English

  12. Why agent network permissions need ongoing checks

    Codex network policy changes explain revocable permits for redirects, long responses, and WebSockets, and why tool and network permissions need separate layers.

    Learning guides Security & privacy Runtime & infrastructure Agent engineering
    English

  13. GitHub Agentic Autofix starts retaining repair experience

    GitHub Agentic Autofix reads and writes Copilot Memory. How security repair knowledge crosses agent workflows, and why verification, scope, and expiry matter.

    Technical articles Memory & context Security & privacy Developer tools
    English

  14. Jev-Code-Reviewer prioritizes large PRs as P0/P1/P2

    Jev-Code-Reviewer prioritizes PR changes, expands P0 by default, and retains the original diff and fallback paths so reviewers see key changes first.

    Projects & practice Developer tools Product design
    English

  15. Agent markets: 85% of the loss comes before negotiation

    Anthropic's book market experiment locates most efficiency loss in preference representation. A look at model choice, preference evals, and market rules.

    Technical articles Multi-agent systems Evaluation AI research
    English

  16. GitHub Security Lab makes fuzzing an agent feedback loop

    GitHub Security Lab Fuzzing Taskflow uses coverage gaps to guide model decisions while deterministic tools execute, measure, and persist results.

    Projects & practice Security & privacy Agent engineering Evaluation
    English

  17. Claude takes a candidate to the lab, not a new CRISPR

    About 950 Claude agents helped identify ART. Human wet-lab experiments confirmed short RNA expression, but its main biological function remains unknown.

    Technical articles Agents Multi-agent systems AI research
    English

  18. Financial markets from zero: stocks, Bitcoin, and risk

    A beginner learning path through stocks, ETFs, indexes, Bitcoin, candlesticks, position sizing, paper trading, backtesting, and tax basics.

    Learning guides Finance
    English

  19. Link CLI lets agents pay without real card numbers

    Stripe Link CLI separates purchase intent from payment permission. Agents request a purchase and receive one-time credentials only after user approval.

    Projects & practice Agents Security & privacy Finance
    English

  20. Debugging agents with traces, spans, and OpenTelemetry

    The GitHub Copilot OpenTelemetry update explains agent traces, spans, metrics, events, content capture boundaries, and a telemetry adapter for Lora PI Kit.

    Learning guides Agent engineering Developer tools Runtime & infrastructure
    English

  21. GPT-6 prompt caching becomes an agent state layer

    GPT-6 adds cache breakpoints, diagnostics, prewarming, and mutable configuration. Long-running agents need stable shared prefixes and cache-hit monitoring.

    Technical articles Memory & context Agent engineering Runtime & infrastructure
    English

  22. LobsterAI: send tasks by phone, run them on your desktop

    LobsterAI connects phone messages to a local desktop agent. The phone handles tasks and approvals while the computer works with files, terminals, and browsers.

    Projects & practice Agents Runtime & infrastructure Product design
    English

  23. Deploying agent backends with Python Workers GA

    Python Workers GA supports web frameworks and agent libraries. WebAssembly package compatibility remains the main migration boundary.

    Technical articles Agents Web development Runtime & infrastructure
    English

  24. Why pinning a Git SHA can still run different code

    Plugin4Shell exposes a supply-chain gap: a pinned SHA records the expected version, but installers must also verify the actual HEAD after checkout.

    Learning guides Security & privacy Developer tools MCP & skills
    English

  25. json-render constrains UI with a component catalog

    Vercel Labs json-render defines components, actions, and data bindings before generating schema-constrained JSON that can be checked and replayed.

    Projects & practice Web development Product design Developer tools
    English

  26. A Minecraft agent separates planning from fast decisions

    This Minecraft agent uses Astra for slow planning, JEV for frequent action selection, and Mineflayer for execution, with logs and scripts to verify runs.

    Projects & practice Agents Agent engineering Evaluation
    English

  27. When to split a task across agents: check dependencies

    How task dependencies shape multi-agent results, with SAIGE experiments, ablations, and OpenAI guidance for deciding when to delegate.

    Learning guides Multi-agent systems Agent engineering Evaluation AI research
    English

  28. Qwen-Image 2.1: native RGBA and commercial licensing

    Qwen-Image 2.1 offers native RGBA, unified generation and editing, up to 10 reference images, and 2K output. Commercial use needs a separate license.

    Technical articles Multimodal systems AI research
    English

  29. How to score a coding agent that finishes 80% but fails

    Android Bench 2.0 keeps pass rate as a hard gate and adds completion rate to diagnose progress and unfinished acceptance requirements.

    Technical articles Evaluation Developer tools AI research
    English

  30. FDE growth: 519 overseas jobs, 28 in China, 6 projects

    A study of 519 overseas job listings, 28 roles at major Chinese firms, and 6 projects examines FDE responsibilities, growth, costs, and scalability.

    Technical articles Agents Developer tools
    English

  31. Jev gives agents typed decisions instead of generated text

    TypeSafe Jev uses Choice, Score, and Noul for constrained agent decisions while code retains control of valid actions and side effects.

    Technical articles Agent engineering Agents
    English

  32. OpenCloak: replace personal data before sending it to AI

    OpenCloak replaces names, emails, and other personal data with stable stand-ins locally, then restores them when the model responds.

    Projects & practice Security & privacy Developer tools
    English

  33. Why you cannot just rerun an agent after a crash

    WorkflowAgent, step boundaries, and idempotency keys explain safe recovery after crashes, timeouts, and human pauses, with a proposed Lora PI Kit adaptation.

    Learning guides Agent engineering Runtime & infrastructure
    English

  34. Anthropic opens internal processes to embedded evaluators

    Anthropic and Accenture extend evaluation beyond model outputs to training, deployment, monitoring, and incident processes, with independence unresolved.

    Technical articles Evaluation Security & privacy AI research
    English

  35. Where agent credentials belong: host, vault, and broker

    Where keys belong when coding agents call GitHub, databases, and cloud APIs, and how vaults and credential brokers define access boundaries.

    Learning guides Security & privacy Agent engineering Runtime & infrastructure
    English

  36. skill-up: evaluate real stateful multi-turn sessions

    Alibaba skill-up resumes real sessions turn by turn, letting skill evaluations test context continuity, intermediate state, and per-turn assertions.

    Technical articles Evaluation MCP & skills Memory & context
    English

  37. Splash: a local LLM inference stack built for each model

    Splash combines model packages, kernels, memory planning, and compatible local APIs into a model-specific inference stack for coding agents.

    Projects & practice Runtime & infrastructure Developer tools
    English

  38. Why more agent tools can mean worse tool selection

    How Tool Search loads definitions on demand, affects prompt caching, and differs from MCP connections and ARD resource discovery.

    Learning guides MCP & skills Memory & context Agent engineering
    English

  39. Claude merges Chat and Cowork to reduce mode choices

    Claude combines Chat and Cowork in one conversation, routes capabilities automatically, and keeps task state and permissions visible.

    Technical articles Agents Product design
    English

  40. Foreman: an independent supervisor for coding agents

    How Foreman separates a coding agent from supervision, using runtime evidence, Jev judgments, and deterministic policy to continue, verify, stop, or escalate.

    Projects & practice Agent engineering Evaluation Developer tools
    English

  41. Gemini Live: turnComplete no longer means the task is done

    Gemini Live Extended Thinking separates conversation turns from background task lifecycles. Clients need interaction_status to track actual completion.

    Technical articles Multimodal systems Agents Runtime & infrastructure
    English

  42. Turn production agent failures into regression tests

    Use Kitaru Replay, Evaluator, and Regression Suite to preserve production failures as repeatable agent tests, with tool policies and CI gates.

    Learning guides Evaluation Agent engineering Developer tools
    English

  43. Why long tasks need more than one growing conversation

    How OpenAI Agents API and Anthropic Managed Agents use sessions, context windows, compaction, and persistent state for long-running agents.

    Learning guides Memory & context Agent engineering Runtime & infrastructure
    English

  44. Skills Manager: sync one skill to multiple coding agents

    Skills Manager distributes a local skill library to Claude Code, Codex, Cursor, and Gemini CLI while managing version and configuration consistency.

    Projects & practice MCP & skills Developer tools
    English

  45. What Tau solves by splitting an AI agent into three layers

    Tau separates model adapters, the agent core, and the application so tools and frontends can work together without depending on each other.

    Learning guides Agents Agent engineering Developer tools
    English

  46. Connecting an agent loop to a frontend with events

    Follow tool execution and repeated model calls, separate streamed deltas from final messages, and explore frontend state with a model-free Python demo.

    Learning guides Agents Agent engineering Web development
    English

  47. AWS agents propose; deterministic code validates

    How AWS separates agent proposals, rule checks, human approval, and execution, with code enforcing boundaries and idempotency.

    Technical articles Agent engineering Multi-agent systems Evaluation
    English

  48. A reliable agent might leave the code alone today

    GitHub Agentic Workflows jsweep checks whether a change is worthwhile and treats noop as valid, instead of forcing automation to produce a diff.

    Technical articles Agent engineering Developer tools
    English

  49. HyperFrames makes video a buildable HTML project

    How HyperFrames gives agent-generated video an inspectable workflow of HTML, animation, assets, lint, preview, and rendering.

    Projects & practice Multimodal systems Web development Developer tools
    English

  50. Herdr turns the terminal into an agent runtime

    How Herdr manages existing coding-agent terminals with persistent sessions, blocked states, cross-machine control, and agent coordination.

    Technical articles Agents Developer tools Runtime & infrastructure
    English

  51. RTK: filter shell output before it reaches a coding agent

    RTK filters test, Git, search, and build noise before command output enters model context, applying context engineering at the tool-output boundary.

    Projects & practice Memory & context Developer tools Agent engineering
    English

  52. Google packages Skills and MCP in one plugin

    How the Google Cloud Developer Plugin packages Skills, MCP configuration, and rules, and what this means for distribution, versions, and permissions.

    Technical articles MCP & skills Developer tools Security & privacy
    English

  53. OpenCompany: version-controlled agent configuration

    OpenCompany uses company.toml, agent teams, and human sign-off to put organizational structure and agent configuration into version-controlled files.

    Projects & practice Multi-agent systems Agent engineering Security & privacy
    English

  54. Agentic search missed an incident. Test the eval too

    The Anthropic incident review shows why automated monitors and LLM graders need coverage, independence, and tests for their own failure modes.

    Technical articles Evaluation Security & privacy AI research
    English

  55. Munim Computer Use lets agents act on desktop controls

    Munim Computer Use reads the accessibility tree and operates desktop controls by element ID, reducing reliance on fragile screenshot-coordinate automation.

    Projects & practice Agents MCP & skills Developer tools
    English

  56. jcode moves coding-agent bottlenecks into the harness

    How jcode handles multi-agent resources, state synchronization, and verification in its harness, beyond the capabilities of the model alone.

    Projects & practice Agent engineering Multi-agent systems Developer tools
    English

  57. OpenAI agent workdays and shifting builder bottlenecks

    OpenAI organization-level data suggests that more coding-agent use shifts bottlenecks to task decomposition, concurrency, harnesses, evaluation, and selection.

    Technical articles Agents Agent engineering Evaluation
    English

  58. Agent demos are easy. The harness is the hard part

    Strands harness-sdk, Harness MCP Server, and HydraFusion show why agent engineering needs control over context, permissions, state, validation, and recovery.

    Technical articles Agent engineering MCP & skills Runtime & infrastructure
    English

  59. Coding agent security is now a supply-chain problem

    Coding agents read and execute skills, MCP tools, dependencies, and repository instructions. These need software supply-chain controls.

    Technical articles Security & privacy MCP & skills Developer tools
    English

  60. GitHub Agentic Workflows: agents inside GitHub Actions

    How GitHub Agentic Workflows puts judgment-based automation inside GitHub Actions while retaining permissions, controlled outputs, approval, and security.

    Projects & practice Agents Developer tools Security & privacy
    English

  61. AI Engineering Harness: verifiable delivery for AI code

    A multi-agent engineering workflow uses isolated worktrees, CI, cold-start review, evidence gates, and layered context to control delivery of AI-generated code.

    Projects & practice Agent engineering Multi-agent systems Developer tools Evaluation
    English

  62. Free Vision Skill: visual evidence for text agents

    VEP/1 turns images into short evidence packets. Provider fallback, caching, and credential storage manage cost and reliability for text agents.

    Projects & practice Multimodal systems MCP & skills Memory & context
    English

  63. MiniMax image and video generation for Hermes Agent

    A Hermes MiniMax plugin with image-01, Hailuo async video tasks, CN/global endpoint fallback, end-to-end tests, and a PyPI release.

    Projects & practice Multimodal systems MCP & skills Developer tools
    English

  64. StepFun image generation for Hermes Agent

    The Hermes StepFun image plugin: directory layout, model routing, API calls, billing endpoints, version fixes, and publishing.

    Projects & practice Multimodal systems MCP & skills Developer tools
    English

  65. Vercel Eve Agent Framework

    A source-backed review of Vercel Eve’s filesystem-first agent model, durable execution, tools, sandboxes, approvals, deployment, and Vercel platform tradeoffs.

    Technical articles Agents Agent engineering Runtime & infrastructure
    English

  66. Loop Engineering Harness: controlling agent iterations

    Break down planning, execution, evaluation, state management, and stopping rules to keep multi-step agent tasks controlled and traceable.

    Learning guides Agent engineering Memory & context Evaluation
    English

  67. Sandbox isolation: boundaries, implementations, and practice

    How sandboxes limit untrusted code access to files, processes, networks, and resources, comparing VMs, containers, user-space kernels, and syscall filters.

    Learning guides Security & privacy Runtime & infrastructure
    English

  68. Building WishLive: A Multi-Agent Runtime

    A 48-hour hackathon build of an event-driven multi-agent runtime with registry, workflow, execution, negotiation, settlement, and observability layers.

    Projects & practice Multi-agent systems Agent engineering Runtime & infrastructure
    English

  69. Building a Full-Stack YouTube Clone with Next.js 15

    A repository-backed walkthrough of NewTube: Next.js 15 structure, tRPC and Drizzle data flow, Mux video pipeline, auth, creator tools, and deployment tradeoffs.

    Projects & practice Web development
    English

Reading list

News notes

16 notes

  1. AI Agent engineering reading list, Sep 30–Oct 1, 2026

    News & reading lists AI agent engineering reading lists
    English

  2. AI Agent engineering reading list, Sep 29–30, 2026

    News & reading lists AI agent engineering reading lists
    English

  3. AI Agent engineering reading list, Sep 28–29, 2026

    News & reading lists AI agent engineering reading lists
    English

  4. AI Agent engineering reading list, Sep 27–28, 2026

    News & reading lists AI agent engineering reading lists
    English

  5. AI Agent engineering reading list, Sep 26–27, 2026

    News & reading lists AI agent engineering reading lists
    English

  6. AI Agent engineering reading list, Sep 25–26, 2026

    News & reading lists AI agent engineering reading lists
    English

  7. AI agent engineering reading list | 2026-09-24 to 09-25

    News & reading lists AI agent engineering reading lists
    English

  8. AI agent engineering reading list | 2026-09-23

    News & reading lists AI agent engineering reading lists
    English

  9. AI agent engineering reading list | 2026-09-22

    News & reading lists AI agent engineering reading lists
    English

  10. AI agent engineering reading list | 2026-09-21

    News & reading lists AI agent engineering reading lists
    English

  11. AI agent engineering reading list | 2026-09-18 to 09-20

    News & reading lists AI agent engineering reading lists
    English

  12. AI agent engineering reading list | 2026-09-18 to 09-19

    News & reading lists AI agent engineering reading lists
    English

  13. AI agent engineering reading list | 2026-09-11 to 09-18

    News & reading lists AI agent engineering reading lists
    English

  14. AI agent engineering reading list | 2026-09-11 to 09-17

    News & reading lists AI agent engineering reading lists
    English

  15. AI agent engineering reading list | 2026-09-11 to 09-16

    News & reading lists AI agent engineering reading lists
    English

  16. AI agent engineering reading list | 2026-09-04 to 09-10

    News & reading lists AI agent engineering reading lists
    English

Other contact methods