TL;DR: Feishu CLI hit 10,307 stars in 51 days with a daily release cadence — 11x faster growth than my own CLI tool md2wechat-skill. Comparing the two projects, I found they're converging on the same direction: AI-native CLI design. The gap is structural, not about effort. Here are three takeaways.

---

§First Reaction to the Numbers

From March 25 to May 15 — 51 days, v1.0.1 to v1.0.31, one release per day. When I saw Feishu CLI's release history, my first thought was: am I doing something wrong?

I also build a Go CLI tool: md2wechat-skill, which handles WeChat Official Account content creation. Started January 11, now at 2,214 stars after 124 days. The growth rate difference: Feishu CLI at ~202 stars/day, mine at ~18/day — roughly 11x apart.

I spent time comparing both projects to understand where that gap comes from.

---

§Feishu CLI's Approach

The project description reads: "built for humans and AI Agents" — AI agents are a first-class user, not an afterthought.

Recent releases (v1.0.22 through v1.0.31, one per day) all ship meaningful features. For example, v1.0.31 added time-based filtering for drive sync, an --exclude flag for auth, and — notably — install now auto-skips interactive prompts in non-TTY environments. That last one is designed specifically for AI agents: when an agent runs a CLI and the tool blocks waiting for a y input, the whole workflow breaks. Building around this from the start signals they're genuinely designing for agentic workflows, not just labeling themselves "AI-ready."

Total: 200+ commands, 20+ AI Skills covering IM, docs, spreadsheets, calendar, and meetings.

---

§My Path with md2wechat-skill

md2wechat-skill handles the WeChat content creation scenario. Over the past few months I did a major refactor — v1.8.0 to v2.0.0 — breaking the layout engine into 43 independent modules, 40+ themes, AI-generated images, and direct push to WeChat draft box.

v2.2.0 shipped May 10 with a memory file feature. Previously, using an AI agent to write WeChat posts required re-declaring writing style every session. Now the tool remembers. I also added support for Claude Code, Codex, and OpenClaw, plus an MCP Server integration.

During the May Day holiday, someone attempted commercial redistribution under the open source license. After discovering this, I spent several days converting the license to Source Available — the reason being simple: I didn't want someone to resell what I built. This created community discussion and real friction. I still think it was worth it over the alternative of watching someone commercialize my work.

---

§11x Growth Gap, Same Direction

The interesting finding: despite an 11x growth gap, both tools are converging toward the same goal.

Feishu CLI says "for humans and AI Agents." My v2.2.0 removes the need to re-declare writing style for AI agents and supports mainstream agentic toolchains. Both tools answer the same question: how do CLI tools work reliably for agents instead of getting stuck on interactive prompts?

CLI tools are structurally well-suited for agents: no popups, scriptable, composable, versionable behavior, predictable output format. GUI tools struggle to provide all of this.

The 11x gap has a clear explanation: Feishu has millions of enterprise users as a foundation, a full-time team, and automated CI/CD. I'm a solo developer with none of that. The gap is structural, not a reflection of effort.

---

§Risk

CLI tools have a real barrier: users need to be comfortable opening a terminal.

Feishu CLI has enterprise IT departments pushing adoption and an official support team. Solo tools depend on developer community word-of-mouth. Applying Feishu CLI's growth logic directly to indie projects will produce misleading conclusions.

---

§Three Takeaways

First: AI-native design must be built in from day one — adding it later is a patch. Feishu CLI's non-interactive mode and standardized output were designed to agent specs from the start. Retrofitted AI compatibility tends to have gaps; agents hit edge cases and workflows break.

Second: Small releases beat saving up for big versions. Daily releases with small scope expose issues early and get feedback fast. md2wechat-skill shipping v2.0.0 through v2.2.0 as incremental releases worked far better than the earlier approach of accumulating changes into one big drop.

Third: Indie tool growth follows different logic than big-company projects. Solo tools live or die by community reputation in specific developer niches. Feishu CLI's growth is enterprise-demand-driven — not a template that maps cleanly to indie tooling.