DEV Community: Fenju Fu

Building Stable, Long-Running Agent Workflows: Why Orchestration Matters

Fenju Fu — Tue, 21 Jul 2026 02:44:57 +0000

In the rapidly evolving landscape of AI development, the focus has shifted from single-model capabilities to complex, multi-agent collaborations. Today's GitHub Trending, featuring projects like jamiepine/voicebox and 1jehuang/jcode, reflects this trend towards multimodal interactions and intelligent execution. However, a critical challenge remains: stability in long-running, complex workflows.

Many AI agents struggle with tasks that require multiple steps, error handling, and context persistence. They often "break" mid-execution, losing progress and requiring manual intervention. This is where robust orchestration becomes not just a feature, but a necessity.

Enter `iflytek/astron-agent`

iflytek/astron-agent is an enterprise-grade Agentic workflow platform designed to address these challenges head-on. It provides a comprehensive environment for building SuperAgents capable of handling complex business processes.

Key Features:

Multi-Agent Collaboration: Seamlessly integrate multiple specialized agents to work together on complex tasks.
Long-Running Task Stability: Implement checkpointing and resume capabilities to ensure tasks can recover from interruptions without losing progress.
Complex Task Decomposition: Break down large, ambiguous tasks into manageable, executable steps with clear dependencies.

The Role of `iflytek/skillhub`

A robust workflow is only as good as the skills it utilizes. iflytek/skillhub acts as the central nervous system for skill management, offering:

Version Control: Manage different versions of skills to ensure compatibility and reproducibility.
RBAC & Audit Logs: Enforce strict access controls and maintain comprehensive audit trails for compliance and security.
Skill Registry: A centralized platform for discovering, publishing, and managing AI skills.

Conclusion

As AI agents become more integral to enterprise operations, the demand for stable, reliable, and orchestratable workflows will only grow. iflytek/astron-agent and iflytek/skillhub provide the infrastructure needed to build these systems, moving beyond experimental prototypes to production-ready solutions.

Explore the ecosystem:

Tags: ai-agent, workflow, enterprise-ai, open-source

Astron Open Source · Bi-Weekly Report (Jun 29 – Jul 12, 2026)

Fenju Fu — Mon, 20 Jul 2026 12:04:11 +0000

🌟 Highlights

Holding strong on the OSS Insight Java trending board: across the last three months of Java repositories, SkillHub #4 · Astron Agent #5 — both firmly in the Top 5.
Two new projects open-sourced (Jul 2): MemFlywheel — a file-native long-term memory layer for Agent Harnesses that makes agents smarter with every run; Domux — a low-latency semantic-parsing model fine-tuned from Gemma-4 that turns natural-language home commands into structured data within 150 ms (98.37% accuracy).
A double release: Astron Agent v1.1.0 (Jul 6) gives the Standard Agent Workflow invocation + Link plugins + Mem0 long-term memory in one release; SkillHub v0.2.13 (Jul 3) polishes review governance, password security, and SUPER_ADMIN protection.
Zero-friction onboarding + two offline events: the Astron Agent open-source edition online playground went live (Jun 30 — try it in the browser, no deployment); HER Hack-Astron #1, the wearables × emotional-companion agent challenge (Jul 6), and the 2nd AI Innovation & Development Conference in Tianjin (Jul 11, with Dong Jiang presenting "Enterprise Agentic AI Application Architecture & Knowledge Engineering") landed back-to-back.

👥 Top 10 Active Community Contributors (Jun 29 – Jul 12, 2026)

Based on message volume in the open-source discussion group (automated accounts such as the Astron Q&A bot and the welcome bot excluded); nicknames anonymized. Thank you to everyone who answered questions, asked questions, and built alongside us 🙌

📦 Release Highlights

🤖 Astron Agent `v1.1.0` (Jul 6) — Workflow + Plugins + Long-Term Memory, all in one

🚀 Workflow invocation inside Agents: Standard Agents can persist workflow capabilities and expose them as runtime tool callbacks, usable from both chat and debug paths, with ownership and space-membership validation (#1487).
✨ Link plugin capability: import Link plugins from the plugin store on the ability page, persist selected tools, and invoke them directly in debug/chat; official/admin-owned plugins are treated as trusted import sources (#1448).
🧠 Mem0 long-term memory: configurable Mem0 memory with app-level isolation, non-blocking writes, and improved Chinese memory preservation and retrieval (#1479).
🔒 Hardened official tool debug-URL validation and trusted-owner handling (#1453); fixed a GitHub code-scanning SSRF alert (#1459); several rounds of CI/remote-test stabilization.
🔧 Upgrade notes: deploy Console/Toolkit changes together to enable workflow invocation; review plugin/workflow/Mem0 settings on existing agents after upgrading. Change scope: 154 files, +6,489/−6,559 lines.
📝 Release: https://github.com/iflytek/astron-agent/releases/tag/v1.1.0

🛡️ SkillHub `v0.2.13` (Jul 3) — Review governance & permission hardening

✨ Promotion review upgrade: pending review cards show richer metadata, with status/reviewed-time sorting and expanded backend/frontend + Playwright coverage (#555).
🔑 Password settings restored: a backend-authoritative canChangePassword capability — local-password users can change passwords again, while OAuth-only accounts show an unavailable state (#554).
🔒 SUPER_ADMIN hardening: non-super-admin users can no longer assign or tamper with the SUPER_ADMIN role (#568).
🐛 Fixed scanner local LLM Base URL propagation while keeping the scan API contract stable, with Kubernetes docs and regression coverage (#565); nested relative links now work in skill-package markdown previews (#557); patched the vulnerable undici frontend dependency (#567).
📝 Release: https://github.com/iflytek/skillhub/releases/tag/v0.2.13 ｜ No breaking changes — safe to upgrade.

🧩 Newly open-sourced: MemFlywheel & Domux (Jul 2)

🧠 MemFlywheel: a file-native long-term memory layer for Agent Harnesses — pre-run recall, post-run extraction, and skill distillation; memories are stored as Markdown, so they are inspectable, Git-diffable, and fully evidence-linked. Works with Pi, Hermes, OpenCode, and other mainstream harnesses → https://github.com/iflytek/memflywheel
⚡ Domux: a low-latency semantic-parsing model fine-tuned from Gemma-4-E2B-it (SFT + GRPO) — parses natural-language smart-home commands into a 7-field structured schema within 150 ms end-to-end, at 98.37% accuracy and 100% format compliance, with no device whitelist required; built for edge deployment → https://github.com/iflytek/domux

⚙️ Astron RPA — Scheduled-task trigger support special

🐛 A user reported nightly scheduled tasks occasionally not firing: the team traced it in the logs to a too-short trigger grace window (a busy machine can miss the 1-second window), shipped a hotfix on the spot (patched trigger.py / base_task.py with a wider grace period + verbose logging), and recommended advanced cron expressions; the fix will be merged back upstream.
📚 Added an AGENTS.md project-context document (#813), aligning with the AI coding-agent ecosystem standard.

💬 Community Q&A & Discussions (Bi-Weekly Picks)

Grouped by product line; all are real community questions + the engineering team's solutions. 🟢 Resolved / 🟡 In progress.

SkillHub (mostly ecosystem integration & governance)

🟡 Could an MCP marketplace be integrated into SkillHub to build a unified enterprise capability platform (Skills + MCP)? Architecturally feasible — SkillHub uses pluggable SPI boundaries and already ships a ClawHub CLI compatibility layer as a foundation; logged as a product-direction request.
🟡 When an agent improves a skill, can it contribute the improvement back to the shared registry as a PR? A great idea for closing the skill co-building loop; the user was invited to file an issue.
🟢 v0.2.13 upgrade notes: no breaking changes; if your Scanner uses a local LLM, mind the Base URL setting (the propagation bug is fixed in this release).

Astron Agent (mostly second-development & new capabilities)

🟢 White-labeling for enterprise redistribution: after changing frontend branding, rebuild the image with each module's Dockerfile under its build directory and swap the official image (same for backend); along the way the team also resolved a Docker image pull failure (switch to a domestic mirror) and a workflow publish failure (fill in the workflow's basic info at the top-left of the canvas first) — the customer demoed successfully the next day.
🟢 Is the platform decoupled from iFLYTEK models? Long since — any OpenAI-protocol-compatible third-party or self-hosted model can be added on the Model Management page.
🟢 Is there an "autonomous planning" agent capability? The platform already supports multi-agent orchestration, workflows, tool/MCP integration, and Agentic Workflow Auto-Generation; file an issue for a standalone autonomous-planning app.
🟢 An intranet WPS document-processing plugin? Not yet — issues/PRs welcome.

Astron RPA (mostly deployment environments & builds)

🟢 Nightly scheduled task not firing: see the special above — grace-window hotfix + verbose-log diagnosis, closed the same day.
🟢 Does it run on cloud desktops / VMs? Yes (the team routinely tests on Tencent cloud desktops); requires Win10 1809 or later; one Windows Server login-hang case is being remote-diagnosed by support 🟡.
🟢 Manual client build fails with Cannot compute electron version: usually pnpm missing from PATH / frontend dependencies not installed — check the environment against BUILD_GUIDE.zh.md (no Java/Docker needed); remote assistance offered.
🟢 How do teams keep script versions consistent without git-style merging? Use the built-in workflow version management (publish/rollback/diff) plus the Center of Excellence as the team's centralized "process market" distributing stable versions.
🟢 Can the open-source RPA be invoked standalone, without the Agent platform? Yes; the cloud RPA has no standalone API yet — call it through the Agent platform and expose the agent as an API, with a video course and a hands-on article to follow.

📌 Q&A entry points: @-mention the Astron Q&A bot in the group (24/7), or ask on the web → https://awesome-astron-workflow.dev/chat

🙌 Community Call: Come Contribute! (Good First Issues & Bounties)

Newcomer-friendly and rewarding from your first commit (contribution points are redeemable at the points store):

🌱 Good First Issues (great for a first contribution)

Project	Issue	What it's about
Astron Agent	#1412	Expand the docs FAQ
Astron Agent	#1410	Contribute a workflow example to the gallery
Astron Agent	#1405	"Are you using Astron Agent?" user-story collection
Astron RPA	#805	"Are you using Astron RPA?"
SkillHub	#528	"Are you using SkillHub?"

💰 Bounties (reward · more challenging, more rewarding)

Project	Issue	What it's about
Astron Agent	#1067	Let OpenClaw trigger Astron Agent to schedule an RPA bot for secure operations
Astron Agent	#1066	Build a "ChatClaw" app on the drag-and-drop canvas with visual fine-tuning
Astron Agent	#709	Support the A2A (agent-to-agent) protocol

👉 Want to claim one? Just comment on the issue. Not sure where to start? Ping us in the discussion group and the engineering team will help break the task down.

✨ Community Spotlight

🌐 Astron Agent open-source online playground is live (Jun 30): zero deployment, zero configuration — try it right in the browser → https://astron-agent-nginx.zeabur.app/
💜 HER Hack-Astron #1｜Wearables × Emotional-Companion Agent Challenge (Jul 6): a dozen-plus teams built agents on site — an AI sleep companion, a health guardian, and "Heart Companion," a retro pocket-watch-style emotional-companion agent (the first submission PR, #1493); see issue #1482 for the event archive.
🎤 2nd AI Innovation & Development Conference, Tianjin (Jul 11): iFLYTEK senior architect Dong Jiang presented "Enterprise Agentic AI Application Architecture & Knowledge Engineering"; the community offered a 15-point ticket-redemption perk.

💬 Got questions, ideas, or want to contribute? Join the discussion group, or head straight to the repos under iflytek-astron to open an Issue / PR.
Let's build a thick, solid open-source ecosystem for enterprise-grade AI Agents together 🚀

Originally published in the iFLYTEK open source community discussions: https://github.com/orgs/iflytek/discussions/6

Building a Trustworthy Skill Registry for AI Agents: Why Versioning and RBAC Matter

Fenju Fu — Mon, 20 Jul 2026 04:28:41 +0000

As AI agents evolve from simple chatbots to complex, multi-step workflows, the way we manage their capabilities becomes critical. The current GitHub Trending landscape shows a clear shift: developers are moving beyond single-agent experiments towards governed, multi-agent ecosystems.

But here's the challenge: How do you manage, version, and secure the "skills" that power these agents?

The Problem: Skill Chaos

In many organizations, AI skills are scattered across codebases, hardcoded into agents, or shared via untracked repositories. This leads to:

Version Conflicts: An agent breaks because a skill was updated without coordination.
Security Risks: Unauthorized users can access or modify critical skills.
No Accountability: When something goes wrong, there's no audit trail.

The Solution: `iflytek/skillhub`

iflytek/skillhub is a self-hosted, open-source skill registry designed specifically for AI agents. It brings DevOps best practices to AI skill management.

Key Features

Version Management: Just like npm or PyPI, skillhub allows you to publish, version, and roll back skills. This ensures that agent workflows remain stable even as skills evolve.
RBAC (Role-Based Access Control): Define who can publish, update, or consume skills. This is crucial for enterprise compliance and security.
Audit Logs: Every action, from skill publication to usage, is logged. This provides full transparency and accountability.
Self-Hosted & Open Source: Deploy it within your network to keep sensitive skill data private.

Integration with `iflytek/astron-agent`

skillhub isn't just a storage system; it's an integral part of the AI agent ecosystem. When paired with iflytek/astron-agent, a powerful agentic workflow platform, you get:

Seamless Skill Discovery: Agents can query skillhub for the right skill based on context.
Dynamic Skill Loading: Agents can load skills at runtime, adapting to new requirements.
End-to-End Governance: From skill creation to agent execution, every step is tracked and controlled.

Conclusion

As AI agents become more central to business operations, governance can't be an afterthought. iflytek/skillhub provides the foundation for building trustworthy, scalable, and secure AI agent ecosystems.

👉 Check it out: https://github.com/iflytek/skillhub
👉 Learn more about astron-agent: https://github.com/iflytek/astron-agent

AI #Agent #OpenSource #DevOps #iFLYTEK

Building Trustworthy RPA: How Astron Combines Automation with Observability

Fenju Fu — Sun, 19 Jul 2026 04:15:20 +0000

In the world of Robotic Process Automation (RPA), efficiency is king. But what good is efficiency if the process is a black box? When a workflow fails, can you pinpoint the exact step and reason?

At iFLYTEK, we believe the future of RPA lies in observability. Introducing iflytek/astron-rpa, an Agent-ready RPA suite designed for enterprise reliability.

Why Observability Matters in RPA

Traditional RPA tools often lack transparency. They execute tasks, but providing a detailed audit trail is an afterthought. Astron RPA flips this script. It’s built with the principle that every action must be traceable.

Full Audit Trails: Log every interaction, decision, and outcome.
Real-time Status Tracking: Monitor workflows as they happen.
Human-in-the-Loop: Seamlessly integrate human oversight into automated processes.

The Power of Combination: RPA + Agent

RPA is the "hands," but it needs a "brain." That’s where iflytek/astron-agent comes in. By combining Astron RPA with Astron Agent, you get:

Intelligent Orchestration: Astron Agent designs the workflow.
Reliable Execution: Astron RPA executes it with precision.
Transparent Feedback: Both tools provide detailed logs for continuous improvement.

This combination ensures that automation is not just fast, but also safe, auditable, and adaptable to complex business needs.

Get Started

Ready to build trustworthy, observable automation?

Explore Astron RPA: https://github.com/iflytek/astron-rpa
Explore Astron Agent: https://github.com/iflytek/astron-agent

Beyond Chatbots: Building the Ultimate AI Assistant with Astron's Cloud-Desktop Synergy

Fenju Fu — Sat, 18 Jul 2026 03:46:48 +0000

The recent surge in popularity of tools like openinterpreter/openinterpreter signals a shift in developer expectations. Users no longer just want AI to generate text; they want AI to act on their behalf. They want to control their operating system, manage files, and automate workflows through natural language.

However, there's a friction point: most local AI agents require deep technical knowledge to configure and maintain.

The Astron Approach: Cloud Intelligence + Desktop Execution

At iFLYTEK Astron, we believe the future of AI assistants lies in synergy. We've designed a dual-engine architecture:

AstronClaw (Cloud AI): The "Brain". It leverages large language models to understand complex intents, plan multi-step workflows, and make high-level decisions.
Loomy (Desktop AI): The "Hands". It runs locally on your machine, capable of interacting with UI elements, files, and applications with precision and security.

Why This Matters

This separation of concerns offers significant advantages:

Security: Sensitive data processing can happen in the cloud, while local execution remains within the user's controlled environment.
Usability: Users don't need to be coding experts. A simple voice or text command is enough.
Reliability: Cloud-based planning ensures robust logic, while local execution guarantees low-latency response.

Get Started: A Practical Tutorial

We've released a comprehensive tutorial to help you build your first cloud-desktop integrated assistant. You'll learn how to:

Connect your desktop environment to AstronClaw.
Design a workflow that uses cloud intelligence to guide desktop actions.
Automate repetitive tasks without writing a single line of code.

Check out the full guide here: https://github.com/iflytek/astronclaw-tutorial

Join the community and start building the AI assistant you've always wanted.

Astron Agent Joins the CNCF Landscape

Fenju Fu — Fri, 17 Jul 2026 07:25:55 +0000

Astron Agent, the enterprise-grade open-source agent platform from iFLYTEK, has recently joined the CNCF Landscape. It is listed under the AI Agent - Workflow Orchestration category of the AI Native landscape, alongside projects such as Dify and Temporal, becoming one of the AI agent building tools officially recognized by CNCF. 🎉

About the CNCF Landscape

The Cloud Native Computing Foundation (CNCF) is a foundation under the Linux Foundation dedicated to advancing cloud-native computing technologies and their ecosystem, and is one of the most influential organizations in the cloud and open-source space.

CNCF launched the Landscape project in 2016 to provide developers and enterprises with a resource map of the cloud-native ecosystem: by categorizing and tracking the projects and products in the ecosystem, it helps users quickly grasp the full picture of the cloud-native stack and find the tools and solutions that fit their scenarios. As AI infrastructure evolves rapidly, CNCF has added an AI Native group to the landscape, covering five categories — AI Agent, Inference, Training, Data, and AI Native Infra — to catalog representative technology projects of the AI-native era.

What This Means

Joining the CNCF Landscape recognizes Astron Agent's project maturity, community vitality, and ecosystem impact within the cloud-native community. When developers and enterprises around the world evaluate AI agent workflow orchestration solutions on the CNCF Landscape, Astron Agent now appears as one of the officially listed candidates.

This is another milestone for the iFLYTEK Astron open-source ecosystem in the international open-source community, following Astron SkillHub joining the Agentic AI Foundation (AAIF). Going forward, Astron Agent will continue to collaborate closely with the cloud-native and agentic AI communities, welcoming more developers and enterprises to build together.

About Astron Agent

Astron Agent is an enterprise-grade, commercial-friendly agentic workflow platform for building next-generation SuperAgents with multi-agent orchestration capabilities. The project provides visual workflow orchestration, MCP tool integration, RPA automation, and enterprise-grade high-availability deployment, is open-sourced under the Apache 2.0 license, and has earned more than 8,600 GitHub stars.

Astron Agent is also a core project of the iFLYTEK Astron Landscape — the world's first enterprise-grade open-source agent project landscape. Presented in the style of the CNCF Landscape, it maps the full iFLYTEK Astron open-source ecosystem, including the agent orchestration platform astron-agent, the RPA automation suite astron-rpa, the official skill library iFly-Skills, and the skill registry SkillHub, covering the complete agentic workflow stack from logic orchestration and tool invocation to skill governance.

Building Resilient AI Workflows: How Astron-Agent Handles Long-Running Tasks with Checkpointing

Fenju Fu — Fri, 17 Jul 2026 03:54:02 +0000

As AI Agents evolve from simple chatbots to complex, long-running processes, a critical challenge emerges: reliability.

When a multi-step workflow runs for hours, involving multiple external APIs, database updates, and conditional logic, the risk of failure increases exponentially. A network timeout, an API rate limit, or a service outage can halt the entire process. Restarting from scratch is not just inefficient; it's often impossible due to side effects (e.g., partial data updates).

The trending discussion around tools like lobehub highlights a shift in developer mindset: Agents are being viewed as "employees" that need to be scheduled, monitored, and managed over long periods. But for enterprise adoption, the core requirement isn't just orchestration—it's resilience.

The Engineering Challenge: State Persistence

In traditional scripting, we might use cron jobs or message queues. But with AI Agents, the "state" is complex: it includes not just variables, but also the context of previous LLM calls, the decision logic of intermediate steps, and the status of external resources.

How do we ensure that if a workflow fails, it can resume from the exact point of failure, without re-executing successful steps or losing context?

The Astron-Agent Approach: Native Checkpointing

iflytek/astron-agent addresses this by treating workflow state as a first-class citizen. It provides native support for:

State Snapshots: At each step of the workflow, the system can take a snapshot of the current state, including variable values, context, and execution history.
Breakpoint Resume: If a failure occurs, the workflow can be restarted from the last known good state, skipping completed steps and re-executing only the failed or pending ones.
Idempotency & Retry Logic: Built-in mechanisms to handle transient failures, ensuring that retries don't cause duplicate side effects.

Why This Matters for Enterprise AI

For scenarios like:

Long-running data pipelines: Processing millions of records with AI-enhanced steps.
Multi-agent coordination: Where one agent's output triggers a series of downstream actions.
Critical business processes: Where downtime or data inconsistency is unacceptable.

The ability to pause, resume, and recover is not a luxury—it's a requirement.

Getting Started

Explore how Astron-Agent can help you build resilient, long-running AI workflows.
🔗 https://github.com/iflytek/astron-agent

AI #Agents #Workflow #Resilience #OpenSource

Beyond Text: How Astron iFly-Skills Empowers Agents with Industrial-Grade Multimodal Capabilities

Fenju Fu — Thu, 16 Jul 2026 03:53:43 +0000

In the current GitHub Trending landscape, a clear pattern is emerging. While general-purpose LLMs remain popular, the real excitement is surrounding vertical skills and multimodal interactions.

Projects like mattpocock/skills (focused on engineering-specific constraints) and moeru-ai/airi (focused on voice and game interaction) are gaining massive traction. Why? Because developers are realizing that text is not enough. To build truly useful agents, AI needs to "see" and "hear" the world, and it needs to do so reliably.

The "Perception" Gap

Many teams we talk to face a common frustration: their Agent can reason perfectly, but it fails when faced with a screenshot or a voice memo. They try to patch this by integrating generic OCR or Speech-to-Text APIs. But in practice, this leads to:

Format Inconsistency: Dealing with messy outputs from free-tier APIs.
Error Handling: Writing complex retry logic for noise or poor audio quality.
Maintenance Overhead: Keeping up with API changes and model updates.

This "perception gap" is where most multimodal agents stall.

The Solution: Industrial-Grade Skills

Instead of building perception from scratch, why not use skills that have been battle-tested at scale?

Introducing iFly-Skills (https://github.com/iflytek/iFly-Skills), an open-source collection of multimodal capabilities from iFLYTEK. These aren't just demo scripts; they are engineering-ready skills designed for high-volume, real-world applications.

Key capabilities include:

🎤 Speech Recognition & Synthesis: High-accuracy STT/TTS with support for various accents and noise environments.
👁️ OCR & Document Processing: Extract structured data from complex images and documents.
🌍 Translation & Proofreading: Context-aware translation and language correction.

Integrating with Astron-Agent

The power of iFly-Skills is unlocked when combined with Astron-Agent (https://github.com/iflytek/astron-agent), an enterprise-grade agentic workflow platform.

Here’s how it works:

Perception: Use iFly-Skills to convert voice notes to text or extract data from images.
Decision: Pass the structured data to Astron-Agent for reasoning and planning.
Action: Execute complex workflows, such as updating a CRM, generating a report, or sending a personalized response.

This separation of concerns allows you to focus on logic and orchestration, while relying on iFly-Skills for robust perception.

Conclusion

The future of AI agents is multimodal. But building multimodal capabilities from scratch is costly and complex. By leveraging open-source, industrial-grade skills like those in iFly-Skills, you can accelerate your development and build agents that truly understand the world.

Tags: ai, opensource, multimodal, agent

Safe Agentic Execution: How Astron RPA Bridges the Gap between AI and Enterprise Workflows

Fenju Fu — Wed, 15 Jul 2026 03:51:18 +0000

As AI agents become more capable, the fear of "hallucinated actions" in enterprise environments is growing. We've seen tools like Dicklesworthstone/destructive_command_guard emerge to act as a safety net, blocking dangerous commands. But relying on post-hoc blocking is reactive.

True enterprise-grade automation requires a proactive architecture: Separation of Concerns between Reasoning and Execution.

The Problem: Agents are Reasoners, not Executors

Agents excel at processing unstructured data, making decisions, and planning. However, they are not designed for high-precision, deterministic execution of repetitive tasks like filling forms, moving files, or updating databases. When agents are forced to do both, reliability suffers.

The Solution: RPA + Agent Synergy

iflytek/astron-rpa provides an Agent-ready RPA suite that bridges this gap. It allows you to:

Orchestrate with Agents: Use iflytek/astron-agent to define complex workflows, make decisions, and trigger actions.
Execute with RPA: Let Astron RPA handle the precise, rule-based execution of those actions in desktop or web environments.
Ensure Safety: Implement fine-grained permissions and audit trails at the RPA layer, ensuring that every automated action is authorized and traceable.

Key Features of Astron RPA

Agent-Ready Design: Seamlessly integrates with AI agents to receive instructions and return results.
Enterprise-Grade Security: Role-Based Access Control (RBAC) and comprehensive audit logs.
Human-in-the-Loop: Easy-to-implement approval steps for critical actions.
Cross-Platform Support: Automate tasks across Windows, Web, and various enterprise applications.

Conclusion

Don't let agents roam freely. Architect your automation strategy with a clear boundary between intelligence and execution. With iflytek/astron-rpa and iflytek/astron-agent, you can build safe, reliable, and scalable AI-driven workflows.

Tags: ai-agent, rpa, automation, enterprise-ai, iflytek

Astron SkillHub Joins the Agentic AI Foundation (AAIF) as an Associate Member

Fenju Fu — Tue, 14 Jul 2026 12:41:27 +0000

Astron SkillHub has officially joined the Agentic AI Foundation (AAIF) as an Associate Member.

The full member list is available on the AAIF website: aaif.io/members

About AAIF

The Agentic AI Foundation (AAIF) is a neutral, open foundation announced by the Linux Foundation in December 2025, dedicated to advancing open source innovation in agentic AI through transparency, collaboration and standardization. The foundation is anchored by three projects — the Model Context Protocol (MCP) donated by Anthropic, goose donated by Block, and AGENTS.md donated by OpenAI — with platinum founding members including AWS, Anthropic, Block, Bloomberg, Cloudflare, Google, Microsoft and OpenAI.

What this means

Joining AAIF recognizes Astron SkillHub's work in agent skill management. Going forward, SkillHub will collaborate with AAIF project communities such as MCP to help advance the open agentic AI ecosystem.

About Astron SkillHub

Astron SkillHub is an enterprise-grade open-source skill registry for AI agents. It helps teams publish, discover and manage reusable skill packages within an organization, with support for self-hosted deployment, semantic versioning, team namespaces, RBAC governance and audit logs. It is licensed under Apache 2.0.

SkillHub is also a core project of the Astron Landscape — an enterprise-grade agentic open source project landscape from iFLYTEK, presenting the full Astron ecosystem in the style of the CNCF Landscape: the agent orchestration platform astron-agent, the RPA automation suite astron-rpa, the official skill library iFly-Skills, and the skill registry SkillHub, covering the full agentic workflow stack from logic orchestration and tool invocation to skill governance.

Enterprise-Grade Agentic Orchestration: Why Structured Context Matters for Multi-Agent Systems

Fenju Fu — Tue, 14 Jul 2026 03:49:41 +0000

The latest GitHub Trending shows a clear shift: developers are moving beyond single-point AI tools towards multi-agent orchestration and structured context.

Projects like Graphify-Labs/graphify demonstrate the power of converting codebases into queryable knowledge graphs. This provides AI agents with a "global memory," solving the context window limitation that plagues complex tasks.

Similarly, HKUDS/Vibe-Trading highlights the demand for vertical, high-value agents that can execute complex, multi-step decisions.

The Astron Approach: SuperAgent Orchestration

At iFLYTEK, we believe the future of AI is not just about smarter models, but better orchestration.

iflytek/astron-agent is an enterprise-grade platform designed for building SuperAgents. It focuses on:

Complex Task Decomposition: Automatically breaking down large, ambiguous tasks into manageable sub-tasks.
Multi-Agent Collaboration: Assigning specialized agents to sub-tasks, enabling parallel execution and specialized expertise.
Structured Context Awareness: Integrating with knowledge graphs and structured data to ensure agents have the necessary global context for accurate decision-making.

Why It Matters

For enterprises, the challenge isn't just building an agent; it's ensuring that agents can work together reliably, securely, and at scale.

astron-agent provides the infrastructure for this. It's not just a tool; it's a platform for building intelligent, collaborative AI systems.

Explore the platform:
👉 iflytek/astron-agent
👉 iflytek/skillhub

ai #agents #opensource #iflytek #devtools

From Command Blocking to Skill Registries: Securing Enterprise AI Agents

Fenju Fu — Mon, 13 Jul 2026 04:22:18 +0000

When destructive_command_guard hit the top of GitHub Trending, it sparked a crucial conversation: How do we safely empower AI Agents with autonomy?

Most solutions focus on blocking dangerous commands after they are generated. But reactive security is not enough for enterprise-grade applications. We need proactive governance.

The Limitation of "Guard" Solutions

Tools like destructive_command_guard act as a firewall, intercepting specific shell or Git commands. This is valuable, but it has limitations:

Reactive: It only stops known-bad patterns.
Siloed: It doesn't manage the capability itself, only its execution.
No Audit: It doesn't track who deployed the skill or why it was used.

The Enterprise Solution: Skill Registries

For enterprises, the real need is a Skill Registry — a centralized hub where all Agent capabilities are registered, versioned, and governed.

Why `iflytek/skillhub`?

iflytek/skillhub is an open-source, self-hosted skill registry designed for AI Agents. It shifts security from blocking to governing.

Key Features:

Version Management: Every skill has a clear version history. Roll back, compare, and track changes effortlessly.
RBAC (Role-Based Access Control): Define who can publish, update, or execute specific skills. Not every developer needs admin rights to production skills.
Audit Logs: Every skill registration, update, and execution is logged. Full transparency for compliance and debugging.
Self-Hosted: Keep your proprietary skills and sensitive data within your own infrastructure.

Combining SkillHub with Astron-Agent

A registry is only as good as the agents that use it. iflytek/astron-agent provides the agentic workflow platform to consume skills from SkillHub.

Register: Publish your skill (e.g., "DataCleaner-v2.1") to SkillHub.
Authorize: Set RBAC rules (e.g., "Only Data Engineers can execute").
Execute: Astron-Agent fetches the approved skill and executes it within a secure workflow.

This combination creates a Secure, Auditable, and Reusable Agent ecosystem.

Conclusion

Don't just block bad actions. Govern good capabilities.

Adopt a skill registry like iflytek/skillhub to bring enterprise-grade security, version control, and auditability to your AI Agents.

🔗 Explore SkillHub: https://github.com/iflytek/skillhub
🔗 Explore Astron-Agent: https://github.com/iflytek/astron-agent

DEV Community: Fenju Fu

Building Stable, Long-Running Agent Workflows: Why Orchestration Matters

Enter iflytek/astron-agent

The Role of iflytek/skillhub

Conclusion

Astron Open Source · Bi-Weekly Report (Jun 29 – Jul 12, 2026)

🌟 Highlights

👥 Top 10 Active Community Contributors (Jun 29 – Jul 12, 2026)

📦 Release Highlights

🤖 Astron Agent v1.1.0 (Jul 6) — Workflow + Plugins + Long-Term Memory, all in one

🛡️ SkillHub v0.2.13 (Jul 3) — Review governance & permission hardening

🧩 Newly open-sourced: MemFlywheel & Domux (Jul 2)

⚙️ Astron RPA — Scheduled-task trigger support special

💬 Community Q&A & Discussions (Bi-Weekly Picks)

SkillHub (mostly ecosystem integration & governance)

Astron Agent (mostly second-development & new capabilities)

Astron RPA (mostly deployment environments & builds)

🙌 Community Call: Come Contribute! (Good First Issues & Bounties)

🌱 Good First Issues (great for a first contribution)

💰 Bounties (reward · more challenging, more rewarding)

✨ Community Spotlight

Building a Trustworthy Skill Registry for AI Agents: Why Versioning and RBAC Matter

The Problem: Skill Chaos

The Solution: iflytek/skillhub

Key Features

Integration with iflytek/astron-agent

Conclusion

AI #Agent #OpenSource #DevOps #iFLYTEK

Building Trustworthy RPA: How Astron Combines Automation with Observability

Why Observability Matters in RPA

The Power of Combination: RPA + Agent

Get Started

Beyond Chatbots: Building the Ultimate AI Assistant with Astron's Cloud-Desktop Synergy

Astron Agent Joins the CNCF Landscape

About the CNCF Landscape

What This Means

About Astron Agent

Links

Building Resilient AI Workflows: How Astron-Agent Handles Long-Running Tasks with Checkpointing

The Engineering Challenge: State Persistence

The Astron-Agent Approach: Native Checkpointing

Why This Matters for Enterprise AI

Getting Started

AI #Agents #Workflow #Resilience #OpenSource

Beyond Text: How Astron iFly-Skills Empowers Agents with Industrial-Grade Multimodal Capabilities

The "Perception" Gap

The Solution: Industrial-Grade Skills

Integrating with Astron-Agent

Conclusion

Safe Agentic Execution: How Astron RPA Bridges the Gap between AI and Enterprise Workflows

The Problem: Agents are Reasoners, not Executors

The Solution: RPA + Agent Synergy

Key Features of Astron RPA

Conclusion

Astron SkillHub Joins the Agentic AI Foundation (AAIF) as an Associate Member

About AAIF

What this means

About Astron SkillHub

Links

Enterprise-Grade Agentic Orchestration: Why Structured Context Matters for Multi-Agent Systems

The Astron Approach: SuperAgent Orchestration

Why It Matters

ai #agents #opensource #iflytek #devtools

From Command Blocking to Skill Registries: Securing Enterprise AI Agents

The Limitation of "Guard" Solutions

The Enterprise Solution: Skill Registries

Why iflytek/skillhub?

Key Features:

Combining SkillHub with Astron-Agent

Conclusion

Enter `iflytek/astron-agent`

The Role of `iflytek/skillhub`

🤖 Astron Agent `v1.1.0` (Jul 6) — Workflow + Plugins + Long-Term Memory, all in one

🛡️ SkillHub `v0.2.13` (Jul 3) — Review governance & permission hardening

The Solution: `iflytek/skillhub`

Integration with `iflytek/astron-agent`

Why `iflytek/skillhub`?