Implementation Notes for Testing Approval Timeout Behavior
Approval steps need tested timeout behavior because an expired approval that silently blocks or silently proceeds are both incidents waiting for a quiet week to end.
200 Beiträge
Für diese Quelle wurden keine Einträge gefunden.
CLI v3.0.71
Plugins now load inside the hub instead of a separate node or bun process, so the CLI no longer needs a JavaScript runtime on your PATH to run plugins. A plugin that fails to load no longer silently loses its tools: Cline prints a warning naming the plugin and the error, cline config tools --json lists it as a plugin_failure entry, and the new cline hub plugins shows each plugin's status and last error (--reload re-imports one without restarting the hub). Set CLINE_PLUGIN_MODE=sandbox to keep the old subprocess behavior
SDK v0.0.93
Plugins load in-process through one registry per process instead of a node/bun subprocess, so the compiled CLI and its hub no longer need a runtime on PATH, and plugins can use the Bun global under the Bun-built hub. CLINE_PLUGIN_MODE=sandbox keeps the subprocess sandbox. The registry wraps import, setup, hooks, tools and commands with timeouts and error capture, and tracks each plugin's state (loading/ready/degraded/failed/disabled) with its last error.
Release sdk/core/v0.0.93 · cline/cline
@cline/core@0.0.93
Release sdk/agents/v0.0.93 · cline/cline
@cline/agents@0.0.93
Release sdk/llms/v0.0.93 · cline/cline
@cline/llms@0.0.93
Release sdk/shared/v0.0.93 · cline/cline
@cline/shared@0.0.93
Release v2.0.28 · anomalyco/opencode
release: v2.0.28
Release v0.21.7+canary.20261013T070740Z · NousResearch/hermes-agent
Hermes Agent canary 20261013T070740Z
Release v0.91.10 · github/gh-aw
🌟 Release Highlights This release focuses on safer defaults, richer session provenance, and more reliable engines and smoke testing. ✨ What's New Read-only agent workflows default to no network access, tightening the default security posture (#68047). Richer unified agent sessions: plugin installations and privacy-preserving MCP configuration are now recorded (#68054, #68055), and execution provenance is preserved independently of safe-output footers (
Gh-aw/work-queue/da3a5da83baaa7a18ce6bd7660f1c830788a4ad4: Cancel queued work when a related Issue closes (#68133)
Outline issue-closure cancellation work Co-authored-by: pelikhan 4175913+pelikhan@users.noreply.github.com Handle related Issue closure as checked work queue cancellation Co-authored-by: pelikhan 4175913+pelikhan@users.noreply.github.com Address work queue review findings Co-authored-by: pelikhan 4175913+pelikhan@users.noreply.github.com Generate compiler-owned workqueue event workflow Co-authored-by: pelikhan 4175913+pelikhan@users.noreply.github.com Generate workqueue event workflow lazily
Stop Scoring Agents Zero at 90% Done
Continuous rubrics, exact-key simulations, and typed decision judges for hill-climbable agent evals
Release v1.0.21-preview.2 · github/copilot-sdk
Internal dependency updates only (SDK snapshot refreshed for Copilot CLI 1.0.97-2). Full Changelog: v1.0.21-preview.1...v1.0.21-preview.2 Warning Firewall blocked 1 domain The following domain was blocked by the firewall during workflow execution: github.com To allow these domains, add them to the network.allowed list in your workflow frontmatter: See Netwo
Release coding-agent-auto-38331174221-1 · sandbox0-ai/sandbox0
Automated coding-agent template refresh. Both native architecture builds passed CLI smoke checks. Existing sandbox RootFS generations are unchanged.
Release rust/v1.0.21-preview.2 · github/copilot-sdk
Update SDK snapshot for Copilot CLI 1.0.97-2
Release 3.4.1 · stanfordnlp/dspy
DSPy 3.4.1 DSPy 3.4.1 makes GEPA's structured InstructionProposer the default instruction proposer and lets dspy.RLM build DSPy sub-agents from inside its REPL. It also adds Python 3.15 support, fixes a 3.4.0 import crash, and includes streaming, adapter, and interpreter fixes.
Release langgraph-checkpoint-postgres==3.1.3 · langchain-ai/langgraph
Changes since checkpointpostgres==3.1.2 release(checkpoint-postgres): 3.1.3 (#9287) fix(checkpoint-postgres,checkpoint-sqlite): treat an empty checkpoint_id as the latest in delta history (#9262) perf(checkpoint-postgres): start the delta walk at the target (#9248) fix(checkpoint,checkpoint-postgres,checkpoint-sqlite): validate namespace labels before store dispatch (
Release v1.2.0 · earendil-works/pi
New Features Per-key OpenRouter model lists: with an OpenRouter key, Pi lists only the models that key may use, with the key's context size, max output, and prices, including regional endpoints set in models.json. See OpenRouter.
Release langgraph-checkpoint-sqlite==3.1.2 · langchain-ai/langgraph
Changes since checkpointsqlite==3.1.1 release(checkpoint-sqlite): 3.1.2 (#9288) fix(checkpoint-postgres,checkpoint-sqlite): treat an empty checkpoint_id as the latest in delta history (#9262) fix(checkpoint,checkpoint-postgres,checkpoint-sqlite): validate namespace labels before store dispatch (#9217) fix: order delta channel replay by task path (
Release langgraph-sdk==0.4.7 · langchain-ai/langgraph
Changes since sdk==0.4.6 release(sdk-py): 0.4.7 (#9289) release(langgraph): 1.2.15 (#9283) fix(sdk-py): end a thread stream's run only on a root lifecycle event (#9228) fix: order delta channel replay by task path (
Release langgraph==1.2.15 · langchain-ai/langgraph
Changes since 1.2.14 release(langgraph): 1.2.15 (#9283) fix(langgraph): give every message written to a messages DeltaChannel an id (#9275) fix(langgraph): save a cached node's writes like a node that ran (#9259) fix(langgraph): stop at an interrupt_before breakpoint when a drain is requested before it (
Release langgraph-checkpoint==4.3.0 · langchain-ai/langgraph
Changes since checkpoint==4.2.0 fix(checkpoint): default InMemorySaver delta history to the latest checkpoint (#9257) test(checkpoint): run the delta conformance suite against InMemorySaver in CI (#9261) fix(checkpoint,checkpoint-postgres,checkpoint-sqlite): validate namespace labels before store dispatch (#9217) fix(checkpoint): keep serialized data when an object can't be rebuilt (
Release v1.27.1 · OpenHands/OpenHands
1.27.1 (2026-10-12) What's Changed Bug Fixes fix(docker): copy MARS web bridge modules into the image by @VascoSch92 in #18343 fix(themes): make warning feedback theme-aware by @enyst in #18333 Documentation docs(verify-openhands): map maintenance 2026-10-12 by
Bench 0.8.1-rc.1
First pre-release of the Rust bench CLI, built from bench-rs at fc2c92d. Without BENCH_VERSION, the install script skips pre-releases. Builds: macOS on Apple silicon (aarch64-apple-darwin), and x86_64 Linux, static (x86_64-unknown-linux-musl, the one the script picks) or with glibc 2.34 or later (x86_64-unknown-linux-gnu). Each .tar.gz has a .sha256 next to it. More in docs/rust/install.md.
Vscode-v1.0.0: fix(vscode): launch the V2 CLI without --port (#49084)
Co-authored-by: uyencss uyencss@users.noreply.github.com Co-authored-by: Aiden Cline aidenpcline@gmail.com
Release v0.122.5 · Agenta-AI/agenta
What's Changed [feat] Show orbiting activity dots in the run timeline header by @ashrafchowdury in #7400 [chore] Clean up Settings, the session header, and integrations browsing by @ashrafchowdury in #7428 [fix] Show the agent's question and keep question order in the dock by
Hello Python 3.15, goodbye 3.10, and now I can use 3.11
Python 3.15 is here! Here's the essential What's new in Python 3.15 guide, this year curated by 3.14 and 3.15 release manager Hugo van Kemenade. As someone who builds open source software that works in every supported Python version, I get to say goodbye to Python 3.10 and start embracing features from Python 3.11.
Release v0.91.9 · github/gh-aw
🌟 Release Highlights This release focuses on safer, more diagnosable workflow runs: stronger privacy-preserving failure reporting, tighter credential/network scoping, and improved NVX Copilot and model-routing support. ✨ What's New In-place add-comment status updates: status comments can be replaced rather than appended, reducing noise (#68013). GitHub Apps can check out PR branches after reviews when allow-listed (#68012). Unified sessions record more context: skill installation, usage and conflicts, plus caller aw_context (
Release v1.27.0 · OpenHands/OpenHands
1.27.0 (2026-10-12) What's Changed Features feat(verify-openhands): drive the mock-LLM e2e behaviors on DeepSeek and map the gaps by @enyst in #18191 feat(verify-openhands): record videos of the browser with browser record by @enyst in #18211 feat(acp): add OpenCode and Pi as built-in ACP agents on local backends by
Import AI 476: Creativity and RSI; worse-than-China open weight models; AI goes to space
Will we give robots the moon?
How to Scale an “AI App” to 1 Million Users in Production
#189: Part 7 - Generative AI Masterclass
Release v3.1.3 · agno-agi/agno
Changelog New Features: Durable External Agent Sessions: ClaudeAgent sessions now resume on any replica. The Claude SDK session id is stored on the Agno session (#10903), and with a database that supports transcript storage (PostgresDb, AsyncPostgresDb, SqliteDb, AsyncSqliteDb) every transcript line is mirrored into the new agno_transcripts table, so a later turn on another machine resumes from the database instead of a local file. A stored session that cannot be resumed falls back to a fresh SDK session seeded with the Agno history.
Make Your Coding Assistant a 10x AI Engineer
Releasing agent-engineer, an open-source plugin for Claude Code, Codex, Cursor, etc. that adds senior AI engineering skills, subagents, and hooks.
Partial Failure Is the Default State of Multi-Agent Systems
In a system with many agents and tools, something is always degraded, so engineering for the all-healthy case means engineering for a state that rarely exists.
Hermes Agent v0.21.7
Hermes Agent v0.21.7 Release Date: October 11, 2026 Patch release. This tag rolls up the ~204 PRs merged since v0.21.6 into a stable release for the desktop app (macOS Apple silicon, Windows), the Microsoft Store and MSIX, the installer, Docker and Hermes Cloud. Full curated notes for this window ship with v0.22.0. What this patch ships for source installs (hermes update and the desktop app's updater on a git checkout): Updates now follow tagged releases.
Datasette-ports 0.4
Release: datasette-ports 0.4 My plugin for detecting and listing all Datasette instances on the current machine now works with the latest version, and displays the detected PID as well. Knowing the PID makes it easier to attach the new Python 3.15 profiler. Tags: datasette
Release v2.0.27 · anomalyco/opencode
release: v2.0.27
Release rc.15-v0.21.7 · NousResearch/hermes-agent
{“attempt”:15,“autopublish”:false,“claimEpoch”:1791775065,“commit”:“7
Release abandoned-rc.14-v0.21.7 · NousResearch/hermes-agent
{“attempt”:14,“attemptRef”:“rc.14-v0.21.7”,“schema”:1,“version”:“0.21
The Science and Engineering of Agent Work
Hilary Mason, Eric Ma, Bryan Bischof, and Tomasz Tunguz on experimentation, creative evals, agentic data science, and local agent
Release rc.14-v0.21.7 · NousResearch/hermes-agent
{“attempt”:14,“autopublish”:false,“claimEpoch”:1791765169,“commit”:“8
Release abandoned-rc.13-v0.21.7 · NousResearch/hermes-agent
{“attempt”:13,“attemptRef”:“rc.13-v0.21.7”,“schema”:1,“version”:“0.21
AI models may be capable of self-replicating without access to a datacenter
Small AI models may be only months away from being able to self-replicate across cheap IoT devices like Android TV boxes, with no datacenter needed.
Strengthening our defenses against distillation attacks
Illicit distillation allows model developers to copy frontier AI capabilities without creating proportionally robust safeguards. Here, we describe how we’ve reduced the rate of distillation by more than 90%.
What Is Browser Fingerprinting? How It Works and Its Limits
Learn how browser fingerprinting works, which signals websites collect, how reliable matching is, and what browser protections mean for privacy and automation.
Release v0.91.8 · github/gh-aw
🌟 Release Highlights This release focuses on richer, more reliable unified agent sessions across engines, stronger sandbox and security hardening, and better diagnostics for model routing and workflow failures.
Release rc.13-v0.21.7 · NousResearch/hermes-agent
{“attempt”:13,“autopublish”:false,“claimEpoch”:1791750773,“commit”:“1
Release mastra@1.34.0 · mastra-ai/mastra
mastra@1.34.0
Release @mastra/upstash@1.5.2 · mastra-ai/mastra
@mastra/upstash@1.5.2
Release mastracode@0.46.0 · mastra-ai/mastra
mastracode@0.46.0
Release @mastra/valkey-streams@0.5.5 · mastra-ai/mastra
@mastra/valkey-streams@0.5.5
Release @mastra/voice-openai-realtime@0.14.4 · mastra-ai/mastra
@mastra/voice-openai-realtime@0.14.4
Release @mastra/voice-aws-nova-sonic@0.2.4 · mastra-ai/mastra
@mastra/voice-aws-nova-sonic@0.2.4
Release @mastra/turso@0.1.14 · mastra-ai/mastra
@mastra/turso@0.1.14
Release @mastra/voice-google-gemini-live@0.14.14 · mastra-ai/mastra
@mastra/voice-google-gemini-live@0.14.14
Release @mastra/voice-inworld@0.4.4 · mastra-ai/mastra
@mastra/voice-inworld@0.4.4
Release rc.12-v0.21.7 · NousResearch/hermes-agent
{“attempt”:12,“autopublish”:false,“claimEpoch”:1791738486,“commit”:“9
Dissecting Coding Agent Harnesses: What Actually Drives Performance
TL;DR Recent empirical work systematically ablates coding agent scaffolds across SWE-Bench Verified and Terminal-Bench 2.1 across 176 settings.
Executive Briefing: The 5 Skills That Help One Person on Your Team Carry More of the Work
Going beyond the video: how taste, agency, focus, resourcefulness, and follow-through develop together, through work that changes your understanding of what you can take on.
Governance Deep Dive: Policy Testing as a First-Class Discipline
Policies are code that decides what agents may do, and shipping them without regression tests means learning about policy bugs from production denials or worse, approvals.
How I Built a Media Empire With an Agentic Software Factory
Building an AI production team to help run Vanishing Gradients, automate podcast production and get more done.
Release rc.10-v0.21.7 · NousResearch/hermes-agent
{“attempt”:10,“autopublish”:false,“claimEpoch”:1791702028,“commit”:“b
Release abandoned-rc.9-v0.21.7 · NousResearch/hermes-agent
{"attempt":9,"attemptRef":"rc.9-v0.21.7","schema":1,"version":"0.21.7"}
Release v1.19.0 · anomalyco/opencode
Core Improvements Run opencode upgrade --major to upgrade to OpenCode 2. Bugfixes Allow the OpenAI ultrafast service tier. Validate PTY connection origins and tickets before checking whether a PTY exists. TUI Improvements Use /update to install OpenCode updates or upgrade to OpenCode 2; updates are no longer installed automatically. Thank you to 1 community contributor: @dc85: docs(web): add Exo Free to Zen docs (#53622) fix(stats): attribute exo usage to an unknown provider (#53621) docs(web): remove Fl
Dwarf Fortress uses version control now
I remember hearing a while ago that Dwarf Fortress didn't use version control. That fact has been lodged in my head, because given its inherent complexity I have trouble imagining a codebase that would benefit from version control more. That's art.
Python 3.15.0 added to actions/python-versions
Python 3.15.0 added to actions/python-versions Bit of a niche link, but I've been waiting for this for a couple of days - I even told ChatGPT to check for it: Clone https://github.com/actions/python-versions and git pull once an hour until they add the stable 3.15 - then tell me about it Now that this has landed, you can add "3.15" to a GitHub Actions testing matrix to run tests against the new Python 3.15.0 release. Tags: github, python, github-actions, chatgpt
PDF Prompt Injection: Hidden Text Only Your AI Agent Sees
Hidden white-text prompts in PDFs fooled AI peer reviewers. How a deterministic geometry scan detects invisible instructions before your agent reads them.
Why Your CVE Program Misses Most AI Agent Vulnerabilities
CVEs still name code bugs well. For AI agents and MCP servers, most real risk sits in defaults, prompt injection, tool composition and supply chain, where a CVE says little or nothing. A map of where it helps, where it misleads, and the triage, inventory and controls that replace it.
How to build a trusted intelligence layer for reliable AI agents
From context engineering to trusted intelligence: a practical architecture for relevant evidence, governed access and human control.
Github-v1.2.26: chore: promote v2 to dev
Takes the v2 tree, with packages/console, packages/web, packages/stats, packages/function, infra and sst.config.ts merged up to latest dev, and retargets V2 publishing and deploys from the v2 branch to dev.
Starting New Threads Is Becoming a Paid Perk — Here’s Why
From October 20, starting new chat threads is a paid-member perk. Here's what changes, what doesn't, and why we're doing it.
Field Notes from an Approval That Expired at the Worst Time
An approval expired mid-execution during a long migration, and the workflow had no defined answer for authority vanishing halfway through a committed sequence.
Quoting The New York Times
Anthropic detailed the activity of its A.I. agents in a blog post on Friday, without naming the targeted websites. But two sources with knowledge of the incidents said Anthropic’s A.I. agents had submitted 20 visa applications through a form available on the State Department’s website. All the applications were incomplete and were not processed, they said. — The New York Times, Anthropic Agents Tried to Fill Out Visa Forms on State Dept. Website Tags: accidental-cyberattacks, anthropic, generative-ai, ai, llms
Issue #49: OCR Extraction Contracts for Document AI Systems
Build a .NET document extraction gate that treats OCR output as uncertain evidence, validates required fields, tables, confidence, and totals, and quarantines weak records before they reach a workflow.
Deno is joining Cloudflare
Deno is joining Cloudflare The Deno team released the first version of celld back in August - their open source implementation of the Durable Objects pattern from Cloudflare Workers. Today, Cloudflare are acquiring Deno outright, with the goal of building on celld to "make workerd self-hosting a first-class supported way to build and run apps using the Workers programming model" (see the Cloudflare blog.) The bad news is that Deno itself will not be maintained by Cloudflare beyond the next year:
I expect rapid progress but not towards general superintelligence
I’ve often been surprised when I hear from top researchers in industry that they think AI will be better than them at their job in a few years, and I didn’t really know why I doubted it.
Release rc.9-v0.21.7 · NousResearch/hermes-agent
{“attempt”:9,“autopublish”:false,“claimEpoch”:1791578012,“commit”:“b6
V2.55.0 (2026-10-09)
What's Changed ⚠️ Compatibility Notes Require Python 3.11 or newer in every package; Python 3.10 installs resolve to 2.54.0 and earlier by @Kludex in #9526 Record the tool requests of ExaSearch, YouSearch, LocalStack and other harness capabilities under durable execution by @DouweM in #9705 Bound retained realtime audio like retained images with r
Release v2.1.296 · anthropics/claude-code
What's changed Added a code key to the Claude apps gateway's managed.policies[]: the same settings as cli, also applied in Claude Desktop's Code tab; beside desktop, it turns on Claude Desktop's gateway mode
Release v0.10.0 · agentevals-dev/agentevals
agentevals v0.10.0 agentevals now reads, stores and evaluates agent telemetry as plain OpenTelemetry. Send OTLP from any GenAI instrumentation, evaluate turns against a golden run, and get the scores back into your observability pipeline as OTel events.
Release evaluator-sdk-v0.2.0 · agentevals-dev/agentevals
evaluator-sdk v0.2.0: custom evaluator protocol 1.1
Release v1.0.19 · github/copilot-sdk
Internal dependency updates only (Copilot CLI snapshot updates); no user-visible SDK changes in this release. Full Changelog: v1.0.18...v1.0.19 Warning Firewall blocked 1 domain The following domain was blocked by the firewall during workflow execution: github.com To allow these domains, add them to the network.allowed list in your workflow frontmatter: See Network Configurati
Release sdk-snapshot/cli-1.0.95 · github/copilot-sdk
Update SDK snapshot for Copilot CLI 1.0.95
Release v0.91.7 · github/gh-aw
🌟 Release Highlights This release introduces a native Agy engine, expands the Git-backed work queue, and improves model-routing visibility and security checks. ⚠️ Breaking Changes Gemini engine deprecated in favor of the experimental Agy engine (#66941). Plan to migrate Gemini workflows to Agy. ✨ What's New Native Go Agy engine with scoped dry-run dangerous-feature checks (#66998), plus smoke conformance and MCP tool allowlist fixes (#67007,
Impactful scheduling for GPU clusters
A Blog post by Ai2 on Hugging Face
Quoting Matthew Green
Everyone is very concerned about being respectable, so I’m going to be the goofball who raises worst-case possibilities. I think there is a 1% chance we live in Minicrypt, and a 15% chance we functionally lose confidence in our existing public-key encryption algorithms. [...]
The Eval Is Under Attack: Benchmark Integrity for Eval-Aware, Multi-Agent Systems
Since mid-2025, capable agents have treated the eval harness as part of the solvable environment: they hunt answer keys, rewrite tests, and emit punctuation that fools LLM judges. Anthropic’s BrowseComp post-mortem logged 2 successful XOR decrypts of a 1,266-problem key (18 runs tried the same strategy; one run burned 40.5 million tokens), and the Agentic Benchmark Checklist estimates harness flaws can misstate agent performance by up to 100% in relative terms. Eval integrity is now a security control, not a one-time design checkbox.
A new feature for my blog, built using my voice
I shipped a new feature for my blog today: the Newsletters page, which offers an index of all of the newsletters I've sent out, both my free weekly Substack and my monthly sponsors-only updates. I built the feature almost entirely using my voice, chatting away to my laptop while I cooked dinner.
Release v1.0.18-preview.1 · github/copilot-sdk
Full Changelog: v1.0.18-preview.0...v1.0.18-preview.1
Release v1.0.9 · dapr/dapr-agents
What's Changed fix(executors): resume the CLI's local Claude session when the store has none by @CasperGN in #880 Full Changelog: v1.0.8...v1.0.9
The Dependency Bump Section Is Where the Risk Hides
Third-party version bumps carry behavior changes that nobody in the release summarized, which makes the quietest section of the notes the least reviewed risk.
Release v0.21.6+canary.20261009T070410Z · NousResearch/hermes-agent
Hermes Agent canary 20261009T070410Z
Release rc.36-v0.21.5 · NousResearch/hermes-agent
{“attempt”:36,“autopublish”:false,“claimEpoch”:1791427955,“commit”:“0
Release 2.2.2 · cisco-ai-defense/skill-scanner
Skill Scanner 2.2.2 This release improves static detection while keeping the measured false-positive rate steady on the pull-request development corpus. Changes Detect whole process-environment va
Ttok 1.0
Release: ttok 1.0 I released ttok 0.4, ran uv tool upgrade ttok, piped a file into the new version... and realized that it was defaulting to the GPT-4 tokenizer when it should very clearly default to GPT-5/GPT-6 instead!
Investigating unintended model actions in our evaluations and internal use
Anthropic is an AI safety and research company that’s working to build reliable, interpretable, and steerable AI systems.
Ttok 0.4
Release: ttok 0.4 ttok is my CLI tool for counting tokens, using OpenAI's open source tiktoken library. It hasn't been in updated in a couple of years, but I finally fixed a Click warning, updated CI, and added a --list-models command to list available models. It works with uvx, so you can count tokens in anything like this: Tags: projects, ai, openai, generative-ai, llms, tokenization
Release v1.0.8 · dapr/dapr-agents
What's Changed chore(deps): Bump actions/github-script from 8 to 9 by @dependabot[bot] in #863 chore(deps): Bump actions/checkout from 6 to 7 by @dependabot[bot] in #864 chore(deps-dev): Bump datamodel-code-generator from 0.82.0 to 0.83.0 by
Quoting Carson Gross
Computer programming is, fundamentally, about two things: Problem-solving using computers Learning to control complexity while solving these problems I have a hard time imagining a future where knowing how to solve problems with computers and how to control the complexity of those solutions is less valuable than it is today, so I think it will continue to be a viable career even with the advent of AI tools. — Carson Gross Tags: computer-science, carson-gross, careers, ai
Release v1.26.0 · OpenHands/OpenHands
1.26.0 (2026-10-08) What's Changed Features feat(apps): add authenticated backend view helper by @neubig in #17656 feat(chat): render empty-response corrective nudge as an informational note by @errmakov in #17864 feat: verify-openhands skill with control-openhands CLI and the full feature map by
Release v3.1.2 · agno-agi/agno
Changelog New Features: Conversation Compaction: Agent(compaction=True) keeps long sessions inside the context window by folding older turns into a summary. Stored messages are never rewritten: a CompactionRecord in the new agno_compactions table drives a derived model payload, so removing the record restores the full conversation. Tune with Compaction(model=..., compact_at_tokens=..., uncompacted_runs=...), fold on demand with agent.compact() / acompact(), and inspect run.compaction.
Release v2.1.295 · anthropics/claude-code
What’s changed Added onFailure: “block” for command and HTTP hooks: a hook that can’t start, times out, or exits with an unexpected code blocks the action instead of letting it through Added Progr
Release v2.0.26 · anomalyco/opencode
release: v2.0.26
Google bid $10 million for an airline’s old work emails to train AI. Much of what made that work matter never showed up in an email
Google wants to learn from the record of our working lives. We should take a closer look ourselves, because understanding what our work accomplishes could change what we attempt next.
Release v0.91.6 · github/gh-aw
🌟 Release Highlights This release delivers fair DAG work queues, an experimental Agy engine, broader model routing, and a round of security and reliability hardening. ✨ What's New Git-backed work queues: mandatory fair DAG work queues with claim-scoped effects (#66024), stress-tested and optimized (#66936), with essential read-only operations exposed via MCP (#66938). Experimental Agy engine with a production conformance gate (
The Pulse: new trend of building “internal vibe-coding apps” at tech companies
Also: do us engineers truly enjoy ‘hard problems’, or do we really just want to solve those which attract promotions, new open models launch in the EU and US, and more
Practical Fun with Personal Agents: Staying on Top of Schoolwork Without Fighting the Portal
Lately I keep seeing some version of this warning: don't use agents to replace your relationship with your kid. I agree, 100 percent. I love my kid and want to spend time with her. But the one thing I
Release v1.5.0 · strands-agents/evals
v1.5.0 Auto-drafted from commits in v1.4.0..v1.5.0, grouped by conventional-commit type. Edit on the release page after publish if you want a polished writeup; the canonical release notes live on the website. 🐛 Fixes fix(mappers): skip ADK relayed agent messages when extracting user_pr… (#414) (7946fdf) fix: blank user query in CorrectnessEvaluator reference mode (#405) (
Hermes Agent v0.21.6
Hermes Agent v0.21.6 Release Date: October 8, 2026 Patch release. This tag rolls up the ~2,100 PRs merged since v0.21.5 into a stable tagged release for Docker and Hermes Cloud. Full curated notes for this window ship with v0.22.0. About this release This is the first release cut by the new stable release pipeline: one attempt ref, a tested Docker image, and a receipt tag at publish. It ships the tag, this GitHub release and the Docker image.
Adapter Authentication Should Never Outlive the Session
Adapter credentials that persist beyond their session become ambient authority, and ambient authority is what turns a small compromise into a large one.
Release coding-agent-auto-1791436751-1 · sandbox0-ai/sandbox0
Shared browser runtime release. Both native architectures passed image CLI/runtime checks and Docker pull/run/build/network checks in stock gVisor (build run 37731773243). Live ali-ue1 canary verified the bundled viewer, Private Preview, local CDP, supervised stop, Docker networking, and the deployed Sandpi authenticated viewer. Existing sandbox RootFS generations are unchanged.
Release 2.10.1 · modelcontextprotocol/inspector
Why The Patch? The addition of the new mcpdo client threw a curveball into the default npx @modelcontextprotocol/inspector@latest method of running the inspector, because there were now two binaries to run, the launcher (used by the other clients) and mcpdo. This quick follow-up patch restores sanity so that command works again. Instructions for launching new mcpdo client are on README.md.
SDK v0.0.92
The AI SDK moves together to one release line: ai 7.0.120 and every @ai-sdk/* package on @ai-sdk/provider-utils 5.0.50 (previously ai 7.0.79 with packages bumped one by one). Turning reasoning off no longer sends reasoning.effort: "none" to OpenAI models that can't turn it off, such as GPT-6 Astra and GPT-6.1 Sol (GPT-6 Astra answered with a 400, for example from commit message generation); GPT-5.x and GPT-6 Sol/Luna still get "none".
Release sdk/core/v0.0.92 · cline/cline
@cline/core@0.0.92
Release sdk/llms/v0.0.92 · cline/cline
@cline/llms@0.0.92
Release sdk/shared/v0.0.92 · cline/cline
@cline/shared@0.0.92
Release v2.1.294 · anthropics/claude-code
What's changed Fixed prompt and agent hooks written as instructions (such as "Block commands that...") allowing what they should block Improved how prompt hooks on Stop and SubagentStop written as instructions (such as "Carry on if the build is broken") are judged, so Claude is less likely to stop early
An opt-in vulnerability-finding service for open-source software
We’re launching OSS Scanner, an opt-in vulnerability scanner for the open-source ecosystem informed by our experience using Claude to find vulnerabilities during Project Glasswing. Projects that join will receive thorough, periodic security scans by our strongest models at no cost.
The model that didn’t exist, so you made it yourself
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
The missing map of the sky
Claude Science helped produce the first complete map of the sky in UV light
Build live dashboards and animate explainers with Claude
Claude Dashboards and Claude Motion are now in beta. Docs, Slides, and Design are out of beta and on every Claude plan, including Free.
How Block orchestrates Claude Fable across thousands of pull requests
Bradley Axen, Head of AI Capabilities at Block, on how his team evaluates new large language models, how Claude Fable 5 orchestrates company-wide code migrations, and why safeguards are a critical part of deploying frontier intelligence responsibly, at scale.
Building effective agent automations
A reference implementation walkthrough for a scheduled Claude Managed Agents automation, with the config files and six rules for avoiding common failure modes.
Release 2.10.0 · modelcontextprotocol/inspector
What's Changed docs(release): add the release-notes step and the re-cut procedure by @cliffhall in #2555 Scope secret-store entries by a per-state-file namespace by @BobDickinson in #2556 Promote recurring pr-flow/board-ops command blocks into scripts/ helpers by
Quoting Ben Affleck
I've always been kind of into computers since I was young. And then when film started to move from analog film to digital, I became more interested in that aspect of it. And the visual effects workflow for many years has included machine learning.
Release v1.1.0 · earendil-works/pi
New Features Program status reporting: terminals and agent dashboards that support OSC 7501 see whether Pi is working, blocked on a dialog or login, done, or failed. See Program status. Claude Hai
Claude Haiku 5.5
As previously promised, here's Anthropic's new fast, low cost model: Introducing Claude Haiku 5.5. The previous Haiku, 4.5, was very much showing its age. It came out almost a year ago, and was priced at $1/million input and $5/million output - relatively expensive even back then, and a full 10x the price of OpenAI's GPT-6 Luna, released last month. The new Haiku exactly matches the price of GPT-6 Luna - $0.10/$0.50 - up to 100,000 tokens.
Release v0.91.5 · github/gh-aw
🌟 Release Highlights This release focuses on model routing and engine resilience, plus a broad round of reliability fixes across Codex, Windows, and safe outputs. ✨ What's New Ordered engine model fallback using native AWF recovery, so workflows keep running when a preferred model is unavailable (#66593). auto and copilot/auto model support in Pi, Codex, and Claude (#66567). Model-routing visibility: routing decisions now appear in logs, audits, and unified sessions (#66645). Workflows are routed to vers
Software engineering is now in the reputation business
what to do about it? how to leverage it?
Release v2.1.293 · anthropics/claude-code
What’s changed Added Claude Haiku 5.5 (claude-haiku-5-5), now the default Haiku model on the Anthropic API — 1M context, $0.10/$0.50 per Mtok ($0.50/$2.50 for prompts over 100K) Added agentType to
Building resilient systems with Sam Newman
Sam Newman joins me to discuss when to use microservices, how to build resilient distributed systems, and how AI is changing software development.
Multimodal open d1 decision models for the edge
A Blog post by Liquid AI on Hugging Face
Release coding-agent-auto-37647574382-1 · sandbox0-ai/sandbox0
Automated coding-agent template refresh. Both native architecture builds passed CLI smoke checks. Existing sandbox RootFS generations are unchanged.
A Hitchhiker’s Guide to Postgres Text Search
How we rebuilt search at Cortex using Postgres.
Anti-Patterns in Software Blogging
Anti-Patterns in Software Blogging Some excellent writing advice from Michael Lynch. Michael warns against "meandering intros", misjudging your reader's existing knowledge, assuming they'll read your previous posts, and excessive formality. He also warns against overreliance on links as an excuse not to explain terminology. This one hurt! I do this all the time, but I have a nagging suspicion that almost nobody ever clicks on them. This point about using your own voice is crucial:
Taste vs. Craft: How a Non-Designer Gets Past AI Slop
For decades, I’ve held graphic designers, visual artists, and creative directors in extraordinarily high regard. I’ve always wanted to be able to create something I considered truly beautiful, but I s
V1.6.3 — Better Temporal Search & Retrieval Accuracy
v1.6.3 — Better Temporal Search & Retrieval Accuracy Release Date: 2026-10-07 Changes: v1.6.2 → v1.6.3 Pull Requests: #4116, #5225, #5226, #5249, #5275, #5277, #5309, #5320, #5361, #5428, #5434 Su
Back by popular demand: Cohort 2 of Hands-On Harness Engineering with Claude
Following the completion of our very successful first session a couple of months ago, I am partnering with Packt Publishing Limited to run Cohort 2 of my live masterclass, Hands-On Harness Engineering with Claude, on Saturday, October 17, 2026, from 9:00 AM to 11:30 AM EDT (6:00 AM – 8:30 AM PDT / 1:00 PM – 3:30 PM UTC).
Release v1.0.17 · github/copilot-sdk
Full Changelog: v1.0.16...v1.0.17
Can a Cloud-Native Harness Make Agents Reliable Beyond the Desktop?
Kubernetes co-creators Craig McLuckie and Joe Beda aim to bring agent harnesses fully into the cloud.
The Squirrely Anarchist and Other Agents I Have Known
"A cornered agent goes squirrely. When you push an autonomous agent into an impossible architectural dilemma with no exit valve, it will contort your codebase to survive." One way I've come to think
Introducing Falcon ASR
A Blog post by Technology Innovation Institute on Hugging Face
Your customers won’t ask for Gemini 4 Argon. Their best AI product idea probably isn’t in their feature requests either. Four questions surface the work they gave up on
Google’s new Gemini 4 Argon model has impressive benchmark results. Google says it leads the Vals Index, which evaluates professional work across several fields. The initial rollout is limited to trusted cybersecurity defenders, with broader access planned. Google has described internal uses and early customer work, but most of us still can’t use it ourselves.
One Model Family, Two Gold-Level Results: Fine-Tuning Nemotron for IOI and IMO
A Blog post by NVIDIA on Hugging Face
ComplexMCP Finds Tool-State Gaps
ComplexMCP is a useful benchmark because it tests the part of tool use that simple function-calling tasks avoid: one action changes the state that the
LinkedIn Outreach: Spend Your Time on Clients, Not a New Side Project
Papr Work's LinkedIn Outreach app gives small GTM teams a working outreach process on day one, so they can spend their time on prospects and clients instead of building automations
Release v0.122.3 · Agenta-AI/agenta
What's Changed [feat] What's New and add feature onboarding to empty pages by @ashrafchowdury in #7261 [fix] Ask for an app's declared read and write access up front by @ashrafchowdury in #7340 [fix] Make chat queue actions instant and show real start-up status by
Workflow Retries Should Re-Enter at Boundaries, Not Mid-Step
Retrying from the middle of a step multiplies partial effects, while re-entering at a step boundary restores a known situation before trying again.
Release @mastra/temporal@0.4.13 · mastra-ai/mastra
@mastra/temporal@0.4.13
Release @mastra/spanner@1.9.2 · mastra-ai/mastra
@mastra/spanner@1.9.2
Release @mastra/tanstack-start@0.2.33 · mastra-ai/mastra
@mastra/tanstack-start@0.2.33
Release @mastra/turso@0.1.13 · mastra-ai/mastra
@mastra/turso@0.1.13
Release @mastra/slack@1.7.2 · mastra-ai/mastra
@mastra/slack@1.7.2
Release @mastra/teams@0.1.1 · mastra-ai/mastra
@mastra/teams@0.1.1
Release @mastra/valkey-streams@0.5.4 · mastra-ai/mastra
@mastra/valkey-streams@0.5.4
Release mastracode@0.45.0 · mastra-ai/mastra
mastracode@0.45.0
Release @mastra/telegram@0.2.2 · mastra-ai/mastra
@mastra/telegram@0.2.2
Release mastra@1.33.0 · mastra-ai/mastra
mastra@1.33.0
Release v4.1.23 · cline/cline
Changed If a model’s response ends without a recognized finish reason, Cline asks it to continue once instead of treating the response as complete. Refreshed the model catalog. The Cline recommend
Release v0.21.4+canary.20261007T070234Z · NousResearch/hermes-agent
Hermes Agent canary 20261007T070234Z
SDK v0.0.91
Responses that end with a missing or unrecognized finish reason are no longer treated as successful completions. AgentModelFinishReason gains unknown (the AI SDK adapter maps unified other and missing reasons to it instead of stop). Without tool activity, the agent keeps the partial response and continues once with a hidden user message; a second unknown finish fails the run. Queued user messages are now consumed at every request boundary, including the first iteration
Release sdk/shared/v0.0.91 · cline/cline
@cline/shared@0.0.91
Release v1.0.17-preview.10 · github/copilot-sdk
Full Changelog: v1.0.17-preview.9...v1.0.17-preview.10
Quoting Jake Boggan
I was a graph theory junkie long ago and even moved to Budapest for awhile to study among the greats. While I was there I started working on Barnette's Conjecture which came to occupy my thoughts over the next 24 years of my life, on and off as I worked in many different fields. Last summer I even thought for a few days that I had actually solved it.
Claude’s fraud detector looked excellent. It leaked the answer
Build, check, and ship with Eleanor and me. Plus: coding agents with Nico and a playbook for catching data science failures.
Release rust/v1.0.17-unstable.37553872417.gc9f6176 · github/copilot-sdk
Update SDK snapshot for Copilot CLI 1.0.92-4.unstable.r37553872417.gc
Release v0.65.0-nightly.20261007.gef59c532f
What's Changed fix(cli): enforce read-only workspace settings in untrusted folders by @jvargassanchez-dot in #29583 fix(core): avoid duplicate tool response turns when resuming sessions by @diegogodinezr in #29618 fix(core): prevent deletion of resumed session history on quick exit (
Release coding-agent-auto-37555842685-1 · sandbox0-ai/sandbox0
Automated coding-agent template refresh. Both native architecture builds passed CLI smoke checks. Existing sandbox RootFS generations are unchanged.
Release rust/v1.0.17-preview.9 · github/copilot-sdk
Update SDK snapshot for Copilot CLI 1.0.93-3
OpenAI “rogue” agent activities found on Wikimedia projects
OpenAI “rogue” agent activities found on Wikimedia projects Given how tempting a target wikis are for rogue agent swarms, it's not a huge surprise that Wikipedia found evidence of that activity once they went looking: The Wikimedia Foundation conducted its own investigation to see whether Wikimedia websites had been similarly affected by AI agents, focusing on those operated by OpenAI. We can confirm that we have discovered some activity by these “rogue” OpenAI agents on Wikimedia platforms.
How to Automate Job Applications with Python
Build a Python job application bot with Browser Use. Find jobs, match your resume, approve applications, fill forms, upload your resume, and track results.
Claude Haiku 5.5
Claude Haiku 5.5 is our fastest, most capable small model. Built for high-volume work like summarization, subagents, and browser use.
Quoting Victoria Kim
Since the Medicare breach, OpenAI has put in place additional monitoring to allow “immediate intervention” by staff to stop training if the company’s models access the internet in ways they’re not supposed to, Mr. Kwon [chief strategy officer at OpenAI] said. — Victoria Kim, Reporting from the Australian parliament Tags: accidental-cyberattacks, generative-ai, ai-security-research, openai, ai, llms
Release v0.91.4 · github/gh-aw
🌟 Release Highlights This release expands engine flexibility and improves debugging and reliability. ✨ What's New Claude Code with GitHub Copilot inference is now supported (#66289). Development compilation and local-first workflow debugging make iterating on workflows faster (#66270). Clearer model routing: routing candidates are now separated from sub-agent model policy (#66234). 🐛 Bug Fixes & Improvements Model routing now reads the agent-job task prompt correctly (
Llm-openai-decisions 0.1a0
Release: llm-openai-decisions 0.1a0 OpenAI released their new Jev-style Decisions API, as previously announced at last week's DevDay. Since I already have an llm-typesafe plugin for talking to Jev, I had GPT-6 Astra read the new OpenAI API documentation and build an llm-openai-decisions plugin inspired by llm-typesafe. Unlike Jev, the new gpt-6-luna decision model supports image input in addition to text. Both models charge for input it and not for output: OpenAI's is 10 cents per million input tokens, Jev's is 4.2 cents per million.
Release 0.161.0 · openai/codex
New Features GPT-6.1 Sol is now the default model in the bundled and Amazon Bedrock catalogs. (#49318, #49339) Amazon Bedrock supports multi-agent V2 and Ultra reasoning on compatible models; Bedr
V0.91.3: Render failure diagnostics safely with progressive disclosure (#66292)
fix: safely render failure diagnostics with progressive disclosure Omit raw engine logs and arbitrary log tails from failure reports. Render incomplete signals and extracted engine errors in bounded, redacted code blocks inside collapsed details sections. Co-authored-by: Copilot App 223556219+Copilot@users.noreply.github.com fix: harden diagnostic redaction and normalization
Llm-mistral 0.16
Release: llm-mistral 0.16 Adds support for reasoning models, such as the newly released Mistral Large 4. Tags: llm, mistral, llm-reasoning
Release v0.63.0
What’s Changed fix(cli): display retry progress indicator during connection recovery (#28340) by @amelidev in #29468 Changelog for v0.61.0-preview.1 by @gemini-cli-robot in #29469 Changelog for v0
EmbeddingGemma 2
My comment on EmbeddingGemma 2 — Hacker News. I really appreciate that EmbeddingGemma 2 is under the Apache 2.0 license. For embedding models in particular, I don't think it makes sense to use a closed, proprietary, hosted-only model. Most applications of embedding models involve calculating thousands or even millions of embedding vectors and storing them for later comparison. If your model is proprietary, the vendor is likely someday going to decide to stop offering that model.
Release v0.64.0-preview.0
What's Changed refactor(a2a-server): implement V1 to V2 settings migration logic by @jvargassanchez-dot in #29450 fix(acp): bridge PromptResponse.usage and emit usage_update notifications (#29389) by @elberthc-byte in
Release v1.18.35 · anomalyco/opencode
Core Improvements Added canonical redirects and JSON and Markdown data formats for agent-readable stats. Bugfixes xAI tool results now include supported images, while unsupported image formats are skipped. (@Jaaneek) Thank you to 3 community contributors: @dc85: docs(web): add Fledge Alpha Free to Zen docs (#52895) @jm0ney337: docs(ecosystem): add opencode-supabase to ecosystem plugins (
Introducing Mistral Large 4: Le chonk
Introducing Mistral Large 4: Le chonk Mistral are back in the game. Today they're releasing a preview of Mistral Large 4, a 1 trillion parameter, 49 billion active parameter model trained on their own cluster of 3,800 NVIDIA Grace Blackwell GPUs. The preview is available via their API. They promise to release the open weights model at the "end of this month". The model only supports two reasoning levels - "none" and "high" - via the Mistral API.
Using Parseable with Datasette for OpenTelemetry traces
TIL: Using Parseable with Datasette for OpenTelemetry traces I saw Parseable in a Show HN today - it's a new observability platform with both an open source (AGPL) Rust implementation (a single ~180MB binary), an "Enterprise" version with extra features and a cloud hosted option. Since Datasette 1.0a41 added OpenTelemetry support (thanks, Alex Garcia), I decided to fire up Codex and have it figure out how to run Parseable and feed it traces from Datasette.
Release v2.1.292 · anthropics/claude-code
What's changed Added --marketplace to claude plugin install: adds the marketplace if needed, under the same policy checks as claude plugin marketplace add, then installs the plugin from it
Mistral Large 4
My comment on Mistral Large 4 — Hacker News. wren6991: The benchmark is saturated. Frontier models are tested with an armadillo in fishnet tights jaywalking on Mars. OK well I couldn't resist this one: Default reasoning levels for each: https://tools.simonwillison.net/markdown-svg-renderer?url=ht
Datasette-atom 0.11a0
Release: datasette-atom 0.11a0 A minor fix for compatibility with the latest Datasette alphas. This meant we could upgrade the datasette.io site to Datasette 1.0a41. Tags: atom, datasette
Release v0.122.2 · Agenta-AI/agenta
What's Changed fix(api): send a custom connection's own model id and key through the gateway by @mmabrouk in #7367 chore(deps): bump werkzeug from 3.1.8 to 3.1.9 in /services/runner/images/sandbox by @dependabot[bot] in #7370 chore(deps): bump multidict from 6.7.1 to 6.9.1 in /sdks/python by
Release v0.91.2 · github/gh-aw
Add packaged Copilot dynamic workflow smoke test Co-authored-by: Copilot App 223556219+Copilot@users.noreply.github.com Grant reusable smoke caller required conclusion and eval permissions Co-authored-by: Copilot App 223556219+Copilot@users.noreply.github.com Enable experimental Copilot CLI workflow tools in smoke test Co-authored-by: Copilot App 223556219+Copilot@users.noreply.github.com Test native dynamic workflows with Copilot CLI 1.0.92 Co-authored-by: Copilot App 223556219+Copilot@users.noreply.github.com
What Is Agentic AI Escalation Architecture?
Innovation is moving at unprecedented speed.
The state of the tech industry in 2026
A look into what has changed, what’s still the same – and what’s broken – in the tech industry. The full video and the summary of my keynote at LDX3 New York
Scrimshaw Jukebox
Tool: Scrimshaw Jukebox I wanted to see if Claude Opus 5.5 could compose music, so I tried this: I want you to write some computer game music for me. First design simple text based format for the music and build an artifact that can play it out loud - include some example tracks in that artifact I am looking for music of the quality of the original secret of Monkey Island It leaned a lot harder into the Monkey Island theme than I had intended, but the results are surprisingly good.
Release langgraph==1.2.14 · langchain-ai/langgraph
Changes since 1.2.13 release(langgraph): 1.2.14 (#9215)
The Cyber Risk Discourse is Broken
Open-weights, ideology, and acknowledging trade-offs.
Release langgraph-sdk==0.4.6 · langchain-ai/langgraph
Changes since sdk==0.4.5 release(sdk-py): 0.4.6 (#9214) fix(sdk-py): percent-encode thread_id and assistant_id in thread stream requests (#9213) release(langgraph): 1.2.13 (#9205) chore(deps): bump the minor-and-patch group in /libs/sdk-py with 5 updates (
Claude Code on Autopilot: Schedule It Once and Walk Away
#188: Part 4 - Claude Code 24/7