testing

Sergio Pino highlights DigitalOcean's new Simulations feature, part of its Agent Learning and Agent Data Services, which lets teams test agents against simulated production conversations, score interactions, and mock costly calls before deployment.

开发者讨论在 AI 代理规模化工作流中如何验证代理自称的“已完成且测试通过”，提出需要外部独立的验证步骤与可留证的调用记录，而非信任代理的自测结果。

Matt Pocock argues that when AI coding agents write bad tests, the fault lies with the codebase structure that permits such tests, not the agent itself.

Anthropic agents were tested on the U.S. State Department's visa application website, raising questions about how AI agents interact with real government systems and the implications for automated task performance in sensitive contexts.

The celld team describes how they use deterministic simulation testing (DST) to explore event orderings and reproduce hard-to-reproduce bugs in their Cloudflare Workers/Durable Objects runtime, including how a seeded simulator controls scheduling, storage faults, and time to verify system invariants.

Tiden 是一款面向编码代理的意图验证工具，可将用户需求记录为可分支合并的规格，并通过在本地环境中运行测试来验证代码变更是否符合预期，支持 Claude Code、Cursor、Codex 和 Copilot。

Two developers building a local-model desktop agent share how their agent leaked a secret and hit a live database because commands inherited the full environment, and they ask the community whether to allowlist or blocklist variables passed to agent-run commands.

A developer shares an update on their multi-agent governance project with 20 agents and hundreds of thousands of test executions, highlighting that proving agents did what they claimed—not getting them to act—has been the core challenge. The latest iteration adds a hash-chained audit ledger, policy checks, human approvals, and evidence verification running in shadow mode before full enforcement.

开发者介绍 FetchSandbox——一个用于安全测试 AI agent 构建的工作流的沙箱环境，无需连接真实账户即可验证状态、回放 webhook 并捕获重复扣费/重复邮件等问题。

A design checklist of five failure cases to test before trusting multi-agent coding workflows: post-review requirement changes, stale worker claims, lost external write responses, approval scope drift, and restart without conversation context. The author frames these as tests to distinguish 'a run ended' from 'this result was accepted against the current task' while building Backplane.

SineFrame M3 is an open-source (Apache-2.0) pytest-based tool for testing MCP servers and verifying the tool calls made by real agents like Claude Code, Codex, OpenCode, and Pi, with trace visualization and CI gating support.

flinch 0.1.0 is released: a Go developer tool that fails pull requests when code covered by a Contract breaks without any Contract test catching it, the third tool from Impatient Programming after exhale and qualm.

Eon's Era generates a complete simulated enterprise with exact ground-truth answers, writing records into Salesforce, Zendesk, Jira, Slack and Gong via MCP and REST interfaces, enabling deterministic evaluation of enterprise agents without human or LLM judges.

Matt Pocock shares that having shipped his AI Coding Crash Course for beginners/intermediates means the upcoming Cohort will be more advanced, including a new /grill-with-docs session covering DDD, scenario testing, and language refinement.

A developer announced Keel, a pure Go desktop UI framework built on Gio that requires no HTML/CSS/JS/WebView, supporting macOS, Windows, and Linux with a component library, packaging tools, and MCP-based agent testing capabilities.

The author argues that Verification Engineering is becoming a key bottleneck and career direction in AI-native organizations, and shares seven practices: using Rust-based toolchains to shorten feedback loops, layering programmatic E2E tests, having agents perform regular real-world inspections, reducing the load on dev/test environments, verifying code after it's merged, verifying that tests can actually catch bugs, and building observability for the verification system itself.

A tweet discussing a testing trick that reveals whether a user is in the gray release (staged rollout) for Claude 5.5 — asking "Who is Tibo" makes the model leak that it's actually 5.5.

Fable 5.5 is reportedly now included in batch routing, with users able to verify their grayscale rollout access by testing a simple prompt about "Tibo the reset guy" on their Claude accounts.

Guillermo Rauch shares Oskar Kwasniewski's announcement of 'e2e', an open-source agentic testing framework for web, mobile, and beyond that mixes deterministic and agentic test APIs, runnable locally or in CI. Rauchg frames it as an example of 'verification-engineering', the future of combining proofs, tests, benchmarks, and linters with AI.

Playwright Skill is a general-purpose Agent Skill that lets coding agents write and run Playwright browser automation scripts on demand, replacing manual page testing with AI-driven test generation, visible browser execution, and automatic screenshot/log capture. It installs via Claude Code, Cursor, Codex, and Gemini CLI.