Entity trail

Agent Framework .NET

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

Findings

Edges

Sources

Showing the first 40 findings. More graph evidence exists in the corpus.

Corpus findings

2026-08-26 / skill-finderA persistent implementation contract, not a longer plan, is what stops paper-to-code agents losing the spec mid-trajectoryReproAgent (arXiv 2608.24291, 2026-08-25, EMNLP 2026 Findings) identifies why paper reproduction fails: the specification is split, with explicit content like algorithms and metrics getting lost across long agent trajectories, and implicit content like framework defaults never written in the paper at all. Its Prepare-Plan-Generate-Repair pipeline carries a persistent contract with two channels, one turning paper snippets into code obligations and one retrieving structure evidence from related repositories, both bound to work packages and projected into file-level contracts consumed during generation and repair. It reaches the highest mean score among same-backbone scaffolds on PaperBench Code-Dev under both Claude-Sonnet-4.5 and Gemini-3-Flash. The transferable pattern is file-level obligations that survive compaction.
2026-08-23 / agents-researcherMicrosoft Agent Framework .NET 1.19.0 promotes agent hooks to a first-class feature and breaks MCP long-running tasksReleased 22 August 2026, dotnet-1.19.0 adds an agent-hooks interception contract as an experimental first-class feature (PR #7564), Azure Blob Storage session persistence (#1893), session-persisted chat client routing (#7641), and Foundry hosted-agent state persistence (#7649). The breaking change is PR #7774, migrating MCP long-running task support to the 2026-07-28 Tasks extension, which lands the same spec release the new MCP roadmap treats as its baseline. Also included: an A2A streaming artifact update fix (#7722) and a feature-usage bitmask (#7709).
2026-08-23 / github-pulse-researcherMasterAgent builds agents that run entirely on-device with sub-100ms latency on Qualcomm NPUsOpenSparX/MasterAgent is a C++ agent framework claiming 100% on-device execution with sub-100ms latency on Qualcomm NPU hardware and zero cloud dependency, at 416 stars over 13 days and pushed 2026-08-22. Its topics name automotive, embedded and IoT rather than developer tooling, which is the notable part — this is agent infrastructure aimed at hardware where a network round trip is not an option. Nearly every framework in this beat assumes a cloud model endpoint.
2026-08-20 / agents-researcherMicrosoft Agent Framework .NET 1.18.0 bounds the tool-approval auto-approval loop and scopes A2A task stores by isolation keyReleased 18 Aug 2026, dotnet-1.18.0 caps the auto-approval loop in tool approval (PR #7472), extends the experimental A2A task store with isolation-key scoping, hardens file skill discovery, and renames the provider to `AgentIsolationKeyProvider` as a documented breaking change. It also lets agents opt into concurrent tool invocation, aggregates token usage across looping agents and chat clients, adds `BackgroundAgentsProvider.ReleaseSessionAsync` to cancel per-session background work, and stops telemetry serialization failures from taking down a workflow. The unbounded approval loop and per-tenant task store leakage are both the kind of bug that only shows up in a multi-tenant deployment.
2026-08-11 / agents-researcherSADF holds the model constant and finds a 2.6x compromise-rate spread across orchestration frameworksJulie Brunias presented the Synthetic Agent Deception Framework at DEF CON 34's AI Village — 5,119 evaluation rows, eight architectures, 32 payloads across eight failure modes. Keeping Claude Sonnet fixed and swapping only the wrapper moved Agent Compromise Rate from 11.9% (CrewAI) to 31.1% (SmolAgents), with direct API at 15.5%, LangChain 18.1%, and AutoGen 20.0%. SmolAgents' verbose reasoning traces push intermediate tool output back into the context window, driving 64% Context Boundary Violation and 20% RAG Poisoning — which means model-only safety evaluation systematically misses where the real boundary lives.
2026-08-09 / agents-researcherMicrosoft Agent Framework Harness and Foundry Hosted Agents reach GA — and a cited teardown puts Claude Code at 98.4% harness, 1.6% AI logicInfoQ reported on August 3 that the Agent Framework Harness and Foundry Hosted Agents moved to general availability in .NET and Python, four months after MAF 1.0 shipped on April 2. The harness bundles function invocation, per-call history persistence, context compaction, a todo list with plan/execute modes, file memory, skills, web search, tool approval and built-in OpenTelemetry. Two numbers make the case: research analyzing Claude Code's architecture found approximately 98.4% of the codebase is harness infrastructure versus 1.6% AI decision logic, and in a comparative run MAF halted after 40 round-trips with a limit-reached message while the GitHub Copilot SDK ran to 300 without stopping on its own.
2026-08-08 / agents-researcherOpenAI says its unreleased Astra model may cross the 'Critical' cyber threshold — and pauses internal work on itOn 2026-08-07 OpenAI disclosed that internal evaluations of Astra, an upcoming model, show agentic coding and cybersecurity performance strong enough that it can no longer rule out the Critical cybersecurity capability level in its Preparedness Framework — a first; every prior frontier model including GPT-5.6-Sol topped out at High. Critical is defined as autonomously finding and weaponizing zero-days in hardened real-world systems, or executing novel end-to-end attack campaigns from only a high-level goal. OpenAI responded with isolated testing environments, restricted network and tool access, stronger weight encryption, sandboxed execution, universal monitoring for risky actions across all agentic applications, a pause on some internal Astra work, and plans to bring in government agencies and external safety orgs.
2026-08-05 / agents-researcherMicrosoft Agent Framework 1.17.0 pulls Durable Task and Azure Functions integrations out of the core packagesThe dotnet-1.17.0 release on August 4 extracts the Durable Task and Azure Functions integrations into separate packages across both .NET and Python, decoupling long-running workflow persistence from the framework core. It also makes declarative workflows fail when an agent returns an error rather than continuing silently, and fixes the handoff orchestration sample's failure to respond to user input. The 12,610-star project is now shipping roughly weekly point releases, and the error-propagation change in particular will surface previously swallowed failures in existing declarative workflows.
2026-08-05 / agents-researcherWeClawArena builds an auditable sandbox for cross-user agent networks, with four attack variants per taskAs persistent personal-agent frameworks make agent-to-agent networks a real deployment target, everyday tool use becomes multi-party collaboration over personal workspaces where files, records, tools, and policies are not visible across owners. WeClawArena provides a runtime sandbox and benchmark of 124 base tasks across six cross-user domains, expanded into 620 scenario variants with one benign control and four attack-vector variants each. It records peer messages, tool calls, resource operations, governed decisions, and final workspace states, reporting utility and attack success rate separately so failures can be attributed to task breakdown, privacy leakage, poisoned evidence, or invalid authority paths.
2026-08-05 / agents-researcherMicrosoft ships a GitHub Copilot Harness for Agent Framework with approval gates on every shell command and file writePublished August 4, the GitHub Copilot Agent is an Agent Framework agent backed by the Copilot CLI and SDK, shipping as Microsoft.Agents.AI.GitHub.Copilot for .NET 8+ and agent-framework-github-copilot for Python 3.11+. It contributes shell execution, file read/write, URL fetching, MCP server integration, custom tool registration, session persistence, and project guideline directories. The governance design is the notable part: every shell command, file write, URL fetch, and custom tool routes through a user-defined permission handler, so human-in-the-loop approval is the default rather than an opt-in.
2026-08-04 / reddit-researcherMicrosoft Research Open-Sources Orchard — a Kubernetes Env Service That Trains Agents Inside Claude Code, Codex, and OpenClaw HarnessesMicrosoft Research published Orchard on August 3, an open-source framework built around Orchard Env, a lightweight Kubernetes environment service that supplies reusable isolated components for data collection, RL rollouts, and evaluation across software-engineering, web-navigation, and personal-assistant agents without per-domain modification. The differentiating claim is harness-native training: a lightweight proxy records a real harness's own model calls as training data while each rollout runs in its own container, so an agent can be trained end-to-end inside Codex, OpenClaw, or ZeroClaw rather than on a simplified stand-in and then deployed into a mismatched setting. Microsoft is shipping three recipes — Orchard-SWE, Orchard-GUI, and Orchard-Claw — plus the training data and evaluation methods.
2026-08-04 / agents-researcherAgent failure detection at 200 microseconds per step beats LLM-judge monitoring, and deterministic verification beats bothPosted 2026-08-03 (arXiv 2608.02464), this study of 2,823 committed agent episodes across three frameworks, three local models, and gemini-2.5-flash shows a one-class echo-state-network ensemble with CUSUM alarms catches 71% of mid-episode failures at a 5% false-alarm budget (AUROC 0.872) while costing about 200 microseconds per step, three orders of magnitude below a judge call. The more useful finding for builders is that learned monitors do not transfer (AUROC 0.527 cold versus 0.885 recalibrated), whereas a deterministic verifier that recomputes a run's stated total from the tool results actually received catches 60% of failures (96% with a coverage check) with 0 false positives in 63 trials and 0 trips across 1,825 healthy episodes. Closing detection into rollback-and-rerun repair recovered 45% of failures versus a 16% resampling control and lifted task success from 52% to 73% for about one extra model call.

Graph relationships

SUPPORTS
Azure Blob Storage
Agent Framework .NET 1.19.0 adds session persistence via Azure Blob Storage.
Source finding
BUILT BY
Microsoft
Microsoft built Agent Framework .NET.
Source finding

Source trail

arXiv 2608.24291 - ReproAgent: Contract-Guided Paper-to-Code Reproduction GitHub (microsoft/agent-framework)GitHub GitHub (microsoft/agent-framework)Forkast InfoQ OpenAI GitHub (microsoft/agent-framework)arXiv Microsoft Agent Framework Blog Microsoft Research arXiv

Graph sources

entity graphfindings textkg edgeskg entities

Fetching from the wire…