2
1 Comment

5 weeks of skill marketplace data: install rank vs retention rank diverging — what's worked for you?

Co-founder check-in. Running TokRepo (skill marketplace for AI coding agents — Claude Code, Cursor pull skills instead of re-prompting).

5 weeks in. Numbers:

  • 2,400+ skill installs across 100+ skills
  • Top 5: Jina Reader (347), whisper.cpp (312), Tabby (265), Mem0 (247), Pal MCP (240)

The split that's widening at week 5: install rank tracks "first impression" — what people try. Retention rank (revisits, time-in-skill) tracks "actually shipped with." They don't match.

Skills that crush install (one viral mention from a creator) don't always keep users. Skills that compound (debug, verify, refactor flows) hit retention without ever hitting top install.

For IH folks running marketplaces, productivity tools, or anything skill-shaped: how do you reconcile this? Bias to discovery and accept the churn? Optimize for top-5 stickiness curve? Surface retention rank publicly to push behavior?

Currently considering a "still-installed-after-2-weeks" sub-leaderboard. Worried it'll confuse new users.

Anyone solved this for a similar marketplace shape? Even partial answers welcome.

on May 11, 2026
  1. 1

    This is a strong marketplace insight because install rank and retention rank are measuring two completely different promises. Install rank shows curiosity and distribution. Retention rank shows real workflow value.

    For AI coding skills, I’d probably avoid treating “most installed” as the main signal for too long. A viral skill can win discovery once, but the skills that get reused during debug, verify, refactor, and shipping loops are probably the ones that define the marketplace long term.

    The “still-installed-after-2-weeks” leaderboard is interesting, but I’d frame it less like a second leaderboard and more like a trust signal: “skills developers actually keep using.” That makes retention feel like quality, not just another metric.

    One thing I’d watch is TokRepo as the brand. It works if the product stays repo/skill focused, but if this becomes a broader agent-skill marketplace or workflow layer for AI coding tools, a cleaner platform name like Xevoa.com could age better.