Skill Calls Skill

I typed this into my terminal:

/gsd:do lets add more blogs with /generate-blog-post skill

/gsd:do is a natural language router. It reads your sentence, matches it to the right command, and hands off. In this case it matched to /generate-blog-post, a skill that knows how to find stories in planning artifacts and draft posts for this site.

The skill searched two project repos, read retrospectives and deviation logs, checked a discard file of rejected ideas, and pitched three stories. I picked two, then asked for a third about the process itself. It drafted all three, self-checked for banned words and em dashes, and generated hero images.

Four layers deep: a sentence, parsed by a router, dispatched to a skill, which orchestrated file reads and artifact mining across repos. None of it was code I wrote for this specific task. The router is generic. The skill is a 170-line markdown file. The artifact search patterns are just file path conventions.

What it looked like from my side

I said what I wanted. The system figured out the how. A few minutes later:

Three drafts ready:
1. "Shipped in a Day" - Qumio v1.0 in 15 hours
2. "A Skill That Writes Blog Posts" - the generate-blog-post skill itself
3. "Skill Calls Skill" - the routing/orchestration moment

Approve all three? Or changes first?

My actual contribution was picking two of three pitches and then asking for a third one about what was happening in real time.

The part that's weird

I didn't plan to write about this. The third pitch was "blog about this prompt," referring to the skill itself. But the real moment was the routing layer above it. The fact that "lets add more blogs" was enough to trigger a multi-step content pipeline.

It's not magic. Each layer is simple on its own. A command parser. A markdown skill file. A file search convention. Stacked together they turn a casual sentence into three finished blog posts with hero images.

This post is one of the three.