AI Agent Artifact Sharing Solved with AWS and CLI: "HTML Share-kun" Released

A tool has emerged that offers a practical solution to the “where do the artifacts go” problem that arises when viewing and sharing AI agent work results. It is “HTML Share-kun,” developed by Qiita user minorun365 and released as open source.

Challenges Facing Conventional Artifact Sharing

Among developers using AI coding agents—particularly Cursor and Claude Code—artifact management has become a new challenge. Practical friction arises: paste results into chat and the conversation moves on; place them in a shared folder and they cannot be accessed from a smartphone while out; and every upload to cloud storage requires permission settings.

The Qiita article points out that this problem stems from the fact that while the “creating” process has accelerated, the conventional workflow of “showing,” “reviewing later,” and “responding on the go” has not caught up. The underlying need is especially to comfortably view visual artifacts generated in HTML format (tables, schedules, Before/After comparisons, etc.) in a smartphone browser.

Basic Design Philosophy and Technical Architecture

The tool’s design philosophy is based on five principles: manage the originals in Git and distribute copies; have agents instructed in Japanese rather than having humans upload manually; flexibly adjust the scope of public exposure according to the recipient; verify completeness on a smartphone; and make the smartphone a bidirectional work window rather than merely a viewing terminal.

Technically, it is realized with an extremely simple configuration. Just place HTML files in AWS S3 and distribute them via CloudFront. No LLM or complex backend is used, and costs are reported to be minimal. A CLI (command-line interface) is provided, ensuring versatility so that it can be invoked in the same way from different AI agents such as Claude Code, Codex, and Cursor.

Operational Refinements and Features

One point the developer particularly refined is eliminating the decision of whether to publish. The original article concludes from actual measurements that AWS costs and token usage are within the margin of error, and that the decision itself is costly. As a result, 273 pages were published within about three weeks of starting to use it.

Another feature is that URL expiration management is stateless. URLs handed to colleagues and editors are signed with HMAC-SHA256, and a CloudFront Function verifies only the signature and the expiration. Because no state such as “this URL is valid” is written to an external database like DynamoDB, it requires no management and automatically becomes inaccessible upon expiration.

Furthermore, an asynchronous workflow between smartphone and PC is also implemented. You can toss a task that comes to mind into an “inbox” from your smartphone while out and pick it up when you return to your PC; conversely, you can request confirmation from the PC side and return approval from your smartphone on the train. This differs from conventional Claude Code remote control, which assumes the Mac is always powered on, in that it operates asynchronously.

Editorial Opinion

Short-Term Impact This tool could immediately affect the daily work of developers and product managers who leverage AI agents. If the cycle of artifact sharing and feedback becomes smooth, the loop of instructing agents and confirming results will turn more easily, likely contributing to productivity across actual operations. In particular, in workflows that use multiple AI agents together, the ability to manage artifacts with a common CLI is noteworthy as a first step toward standardization. Long-Term Perspective In the long term, this could become a foundation that changes the very “way” AI agents are used. If artifacts come to be managed not as mere chat logs but as assets that can be comfortably accessed and operated from a smartphone, the quality and quantity of work “delegated” to agents will expand. Moreover, as lightweight foundations like this tool become widespread, AI agent developers will likely advance designs that are more conscious of the “user-side experience” of how artifacts are utilized. Question from the Editorial Desk This case offers an important lesson: the “delivery and management” infrastructure for artifacts produced by AI agents is directly connected to human productivity.

References

• “AIエージェントの作業結果、マークダウンで読むの辛くない？ →「HTML共有くん」を作りました”, by minorun365 — Qiita, 2026-08-16T06:57:25.000Z (ARR)
• Source URL: https://qiita.com/minorun365/items/320b4230c0b0c169ba13?utm_campaign=popular_items&utm_medium=feed&utm_source=popular_items