6 months in, the hardest part of TokRepo isn't building
Co-founding TokRepo (skill marketplace for AI coding agents), no PM hands you the roadmap. Every week we re-bet what's worth shipping next.
This week's check: what skills are users actually pulling?
Top 5 over the last 4 weeks:
- Jina Reader fetcher — 347 installs
- whisper.cpp pipeline — 312
- Tabby — 265
- Mem0 — 247
- Pal MCP — 240
All five are debug / verify / observability skills. Not a single "ship faster" skill in the top 10. ~60% of install volume comes from skills under 200 lines.
Three months ago we'd have built a "10x faster scaffolding" onboarding because that's what every competitor pitches. Looking at the data, that's the wrong fight. People extracting depth want to codify what they already understand and compose it — not skip the understanding.
So the next bet: better skill versioning + diffs, less "AI generates this for you" magic.
The "figuring out what to do" muscle is bigger than the building one. Half the job is killing roadmap items the data tells you not to build.
