A SkillsMP entry for running a Discord-backed OpenClaw session. It is aimed at talking to the active agent rather than searching old Discord messages.

Why it matters: Useful if a client community needs a conversational front door to an already supervised agent.

SkillsMP Messaging Client ops

A testing skill for OpenClaw Control UI changes using Vitest and Playwright, including mocked gateway flows, screenshots, and browser-verifiable evidence.

Why it matters: A good reference for insisting on proof from agent-driven product changes before a client sees them.

SkillsMP Testing Reliability

A release-automation skill for OpenClaw nightlies that uses isolated branches, release CI, retained branches, and a forward-port path back to main.

Why it matters: Useful as a checklist for agent-maintained releases where experiments should not destabilize the working branch.

SkillsMP Release ops Engineering

A memory-maintenance skill for OpenClaw that keeps its knowledge base in predictable pages, tracks managed sections, and ties changes back to evidence.

Why it matters: Worth borrowing if your agent memory needs a reviewable source of truth rather than an opaque context pile.

SkillsMP Memory Knowledge ops

A newly released MCP that checks a business website for a contact route and tests whether that route works. It is directly relevant to prospect enrichment, not generic web search.

Why it matters: Run it against a small lead batch and compare valid-contact yield with your current enrichment chain.

Official MCP Registry Lead research Vesara

ABMeter exposes A/B experiment management through an assistant, with SDK support across Python, JavaScript, Ruby, Go, React Native, and Node.

Why it matters: It could give agents a disciplined way to propose and record landing-page or outreach tests.

Official MCP Registry Experimentation Growth

Project Desk is a newly registered work tracker intended for people directing AI across several projects. The agent can update tasks, context, and progress as work moves.

Why it matters: A candidate for the gap between agent execution logs and the operator view of what is actually moving.

Official MCP Registry Project management Operator ops

Adtest scores image, video, and text ads across 13 dimensions before spend. It is designed for assistant-driven review rather than autonomous ad buying.

Why it matters: Use it as a preflight opinion in a creative workflow, then compare its calls with real campaign results.

Official MCP Registry Creative QA Marketing

Meta introduced Muse as a personal agent with access ambitions that include email, calendars, payments, and health services. The product question is whether users will grant that scope to Meta.

Inception Labs released Mercury 2.5. For operators, the useful follow-up is concrete: benchmark latency, cost, and tool-use reliability on a real production task before moving workloads.

Deltafin demonstrates Kimi K3 running at one token per second on a MacBook Pro by streaming from four SSDs. It is a reminder that storage and serving tricks keep changing the local-model cost curve.

A new paper on HN reports that adaptive exploration can produce novel social biases in language models. Agent teams using self-directed exploration need evaluation checks beyond task success.

Every published a fresh account of onboarding an AI project manager. The useful operator angle is the ongoing maintenance work: delegation creates another system that needs clear ownership and review.

A builder adapted Lean manufacturing ideas so recurring Claude Code mistakes become tracked failure modes. The interesting part is the feedback loop, not the prompt template.

A cautionary account of a wiki-reading assistant surfacing details from a restricted spreadsheet. Retrieval permissions and answer-time policy checks need to be designed together.

Raggy combines vector search and BM25 over local files, with local or remote generation. It is a compact option for testing document retrieval without starting with a hosted stack.

A practitioner asks how to make a local coding agent stop immediately without letting it alter its own guardrails. That is a real design requirement for unattended execution.

Mistral confirmed a €3 billion Series D at a €21 billion valuation. Sovereign AI remains a major capital story in Europe, with infrastructure and regional control part of the pitch.

Paris-based Actionable raised €8.6 million for predictive customer-experience software. Its product focuses on telling enterprises who may churn, complain, or buy again and why.

The authors propose diffusion-augmented language models to reduce the sequential bottleneck in text generation. Strong operator fit: serving speed and cost are still constraints on agent throughput.

NeoHorse-1 explores agent-native post-training with a routing harness that turns observed performance into training signals. Medium fit today, but relevant to how agent systems may improve from their own runs.

FlowBalance studies self-improvement for reasoning models using terminal verifiers alongside denser guidance. Strong fit for anyone building evaluation loops where confident wrong answers are expensive.

This paper investigates why a hybrid 27B model retains quality under four-bit quantization. It is a practical read for teams weighing cheaper local inference against degradation risk.