🚀 Astron Open Source · Bi-Weekly Report (Jul 27 – Aug 9, 2026)

[FenjuFu]

🚀 Astron Open Source · Bi-Weekly Report (Jul 27 – Aug 9, 2026)

---

🌟 Highlights

• Astron Agent climbed to # 4 on OSS Insight's three-month Java trending board, with 106 stars, 4 forks, 57 pushes, and 4 PRs in the ranking snapshot. SkillHub ranked # 8, recording 46 stars, 19 forks, 92 pushes, and 4 PRs.
• Three releases landed during the cycle: Astron Agent v1.1.1, SkillHub v0.2.15, and SkillHub v0.2.16.
• Across the reporting window, the three core repositories merged 82 PRs: Astron Agent 24 · SkillHub 51 · Astron RPA 7.
• Astron Agent v1.1.1 expanded MCP interoperability with Streamable HTTP plus SSE fallback, while strengthening workflow history, CoT streaming, knowledge parsing, and debug context handling.
• SkillHub moved deeper into production operations with an official Helm deployment path, Redis Cluster/Sentinel support, reverse-proxy sub-path deployment, reviewed built-in Skills, and stronger observability.
• HER Hack-Astron # 2 opened a ¥500 Langfuse observability bounty, inviting contributors to add hierarchical traces, cost/latency monitoring, and evaluation to Astron Agent.

👥 Top 10 Active Community Contributors

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

---

📦 Release & Development Highlights

🧰 SkillHub v0.2.15 and v0.2.16 — production deployment, built-in Skills, and observability

SkillHub merged 51 PRs during the two weeks and published two releases.

v0.2.15 (Jul 30)

• ☸️ Official Helm Chart: a production-ready Kubernetes path now covers PostgreSQL and Redis dependencies, external services, Ingress/TLS, S3 storage, autoscaling, disruption budgets, upgrade safeguards, and OCI publication through GHCR (# 445).
• 🧱 Redis deployment choices: operators can use standalone Redis, Redis Cluster, or Redis Sentinel without changing the default standalone experience (# 505).
• 🔐 More reliable authentication: OAuth Device Flow works with production Redis serialization; invalid API-token responses now include safer reasons and request IDs; incomplete OAuth providers stay hidden; stale login bundles recover after logout (# 607, # 610, # 480, # 560).
• 📦 Publishing and discovery fixes: rejected versions can be uploaded again without losing published/yanked history, and label changes trigger asynchronous search-index rebuilding (# 601, # 592).
• 🧭 Safer CLI coordinates: namespace/slug, @namespace/slug, and namespace--slug are normalized consistently, preventing namespace conflicts and accidental removal of a same-slug install elsewhere (# 608). This change was merged to source but was not yet included in npm CLI 0.1.9 at release time.
• 🐘 Kubernetes PostgreSQL initialization now works when a fresh volume root contains lost+found, while remaining compatible with existing root-level clusters (# 367).

v0.2.16 (Aug 7)

• 🌐 Reverse-proxy sub-path deployment: the web application can run safely under a base path such as /skillhub/, including deep links, static assets, API calls, and authentication redirects (# 576).
• 📚 Reviewed starter Skills: a built-in collection now ships with deterministic package validation and runtime bootstrap verification (# 635, # 636).
• 🔭 Operational observability: request/message correlation, selectable tracing behavior, safer logging, and visible asynchronous index-rebuild failures make production diagnosis easier (# 664, # 684).
• 🐛 Web and registry reliability: fixes cover React 19 overlay teardown crashes, label-to-search navigation, CJK-safe ClawHub download redirects, actionable 4xx responses, and safer skill-version deletion (# 624, # 625, # 674, # 685, # 641).

Welcome to the ten contributors highlighted as first-time contributors across the two release notes: @zhukai9527, @gale-popai, @shychee, @yaffir, @jangrui, @mosesyu95, @xring, @Phil-OSophy-42, @michael-xiii, and @wowo-zZ 🙌

🤖 Astron Agent v1.1.1 — modern MCP transport and a more resilient runtime

Astron Agent merged 24 PRs and published v1.1.1 on Aug 7.

• 🔌 MCP Streamable HTTP with safe SSE fallback: auto mode tries Streamable HTTP first, falling back only for transport negotiation or connection failures—not for authentication, TLS, or post-initialization errors—so a tool call cannot be sent twice across transports (# 1552).
• 🧩 Workflow capability inside Agents is included in the release baseline, letting standard Agents persist and invoke workflows with ownership and space-membership validation (# 1487).
• 🧠 CoT and streaming correctness improved across mixed reasoning/content chunks, usage-only chunks, and duplicate-call prevention (# 1581, # 1582, # 1589).
• 🛡️ Workflow state handling became safer: blank workflow history can no longer poison later runs, SSE callbacks are isolated, disconnects no longer replay requests, and large LLM protocols persist safely (# 1579, # 1585, # 1584, # 1595).
• 📚 RAGFlow ingestion reliability: parsing jobs now have terminal-state protection and timeout/retry safeguards; parser configuration is applied before ingestion, and large-document storage and regression coverage were expanded (# 1597, # 1602).
• 🌏 Documentation discovery improved as the Japanese locale became available in the VitePress language switcher (# 1550).

⚙️ Astron RPA — team-market fixes, easier open-source builds, and clearer integration docs

Astron RPA merged 7 PRs during the reporting window:

• 🐛 Team-market invite and application context fixes restored the invite modal and preserved the selected application when opening team-market dialogs (# 830, # 832).
• 🧱 Open-source Maven builds are now the default for rpa-auth, reducing accidental reliance on non-open profiles during local builds (# 833).
• 📚 Community troubleshooting moved into the repository: cloud-desktop support, client build failures, scheduled tasks, and Agent ↔ RPA connectivity now have clearer guidance (# 824, # 829).
• 📈 Project discovery improved with a Trendshift badge and a clearer star/support call to action (# 827, # 825).

---

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

The following themes were distilled from anonymized WeCom support and open-source discussion records for Jul 27–Aug 9. Automated Q&A/welcome accounts were excluded from the review. 🟢 Resolved · ℹ️ Clarified · 🟡 Follow-up needed.

Astron Agent

• 🟢 Why does a nested workflow fail even though it appears in the editor? Publish the child workflow first and complete its required category/basic metadata. A draft or incompletely published child cannot be invoked reliably by the parent workflow.
• ℹ️ How should a workflow API receive a DOCX/PDF attachment? Upload the file to storage reachable from the deployed Astron Agent environment, then pass the returned URL as the workflow input. File upload and file-content parsing are separate steps: use a plugin or a dedicated parsing service when the code sandbox cannot fetch the URL or lacks the required parser packages.
• 🟢 Why can RAGFlow parsing remain pending or fail on large documents? The cycle landed timeout/terminal-state protection and parser-configuration fixes in # 1597 and # 1602. Upgrade to a build containing those changes and keep the parsing logs when reporting any remaining case.
• ℹ️ Why does a third-party plugin see Chinese text as \uXXXX sequences? JSON may escape Unicode characters while the payload is still transported as UTF-8. The receiving service should parse the JSON body instead of treating the serialized representation as final display text.
• 🟡 How should multiple knowledge-retrieval results be processed one by one? Normalize the retrieval output into a list, then feed it into an iteration branch; for fixed-size emergency workarounds, a code/text-processing node can split the list into explicit variables. Missing filename/title metadata or an iteration shape the current nodes cannot express should be captured in a reproducible issue.

SkillHub

• 🟢 Why does a Skill upload report malformed input? Check both layers: valid YAML frontmatter in the package-root SKILL.md, and UTF-8-safe archive filenames/paths. A YAML error and a non-UTF-8 filename can surface similarly during validation or extraction.
• ℹ️ Can SkillHub sit behind a reverse proxy at /skillhub/? Yes, beginning with v0.2.16. Configure SKILLHUB_WEB_BASE_PATH together with matching public/API base paths so deep links, assets, API calls, and auth redirects agree.
• ℹ️ What about enterprise identity providers? This cycle merged a unified identity core and adapters for CAS, LDAP/Active Directory, and DingTalk, plus secure account linking (# 643, # 657, # 672, # 677, # 668). Operators should follow the documentation and a tagged release that explicitly exposes the provider they intend to use rather than assuming every merged adapter is enabled in an older deployment.
• ℹ️ Why can the CLI remove the wrong same-slug Skill? Upgrade to a CLI release containing # 608. The change was present in source at v0.2.15 time but not yet in npm CLI 0.1.9.

Astron RPA

• 🟢 Why is the team-market invitation dialog blank or missing context? Fixes # 830 and # 832 repaired the modal and preserved the selected application context.
• ℹ️ I deployed the server—where is the desktop UI? The open-source stack includes server and client components, but deploying the services does not itself install the desktop client. Build the client following the repository guide or use the published client download, then point its configuration to the reachable server endpoint.
• ℹ️ Can the open-source project be customized? Yes. The repository contains the open-source client/server code intended for secondary development; compare the open-source architecture and build guide with enterprise-only features before planning a rollout.

---

🔭 Community Spotlight: HER Hack-Astron # 2

Astron Agent opened a new community challenge focused on Langfuse observability for multi-agent and workflow execution.

Contributors can take one or combine several approaches:

• instrument workflow nodes directly with the Langfuse SDK;
• add extensible callbacks to the workflow runner;
• bridge the existing OpenTelemetry foundation to Langfuse through OTLP and semantic conventions.

The winning contribution receives ¥500 RMB plus the HER Hack-Astron # 2 Champion title; additional accepted contributions may receive Astron merchandise, Loomy credits, or coffee vouchers. Submissions should include real trace screenshots/video or reproducible logs, disclose AI assistance as requested by the repository, and link the PR with Closes # 1575.

👉 Claim or discuss the challenge in issue # 1575

---

🙌 Build With Us

Good starting points from this cycle include:

• reproduce a community-reported workflow, upload, knowledge, or RPA build problem and turn it into a minimal test;
• improve bilingual deployment and troubleshooting documentation;
• validate SkillHub behind a real enterprise reverse proxy or identity provider;
• contribute a reusable workflow, Skill, RPA process, or observability integration;
• join HER Hack-Astron # 2 and help make SuperAgent execution easier to inspect and evaluate.

Before starting, leave a comment on the relevant issue so maintainers can confirm scope and avoid duplicated work.

---

🔗 Explore the Astron Open-Source Ecosystem

• Astron Agent — enterprise-grade agentic workflow platform
• SkillHub — self-hosted registry and governance platform for Agent Skills
• Astron RPA — Agent-ready RPA for desktop and web automation
• All repositories tagged iflytek-astron

Methodology note: PR totals use GitHub merged_at timestamps from Jul 27 through Aug 9, 2026. OSS Insight figures come from the supplied Aug 11 snapshot for Java repositories over the past three months. Community questions were anonymized and condensed; customer/group identities and environment-specific details were not published.

Questions, ideas, or a contribution to share? Join the community discussion or open an Issue / PR in the relevant repository. Let's keep building open, production-ready infrastructure for enterprise AI Agents 🚀