@borghei
The whole premise of Claude-Skills is that AI skills belong to every team, not just engineering. But the activation paths — npx @borghei/claude-skills add, Claude Code plugin, manual install — still filter for developers. A marketing lead who wants to run gtm-strategy or a compliance officer who needs soc2-readiness hits a terminal wall before seeing any output.
I set up an online entry point for the library. Anyone landing on the repo can upload their project docs or deliverable drafts, describe the professional task, and get the skill's output directly:
The design decision that stood out: each skill ships with a worked example, an anti-pattern library, and a deterministic scoring rubric. That structure means the online version doesn't just run prompts — it carries the same quality guardrails into every session. A first-time user gets the rubric-scored output, not a raw LLM response.
Visitors can try any skill from the library here — upload a doc, describe the task, get the deliverable. Every run leaves a usage record on the platform, so you can see what tasks people actually bring and where the skill instructions land well or miss.
That signal loop — real inputs from non-dev users running skills on their own work — is hard to get from GitHub stars or CLI installs alone. It's the kind of feedback that sharpens SKILL.md files against actual use patterns.
shesonglin@tinkerland.ai
Feel free to close if this isn't relevant.
@borghei
The whole premise of Claude-Skills is that AI skills belong to every team, not just engineering. But the activation paths —
npx @borghei/claude-skills add, Claude Code plugin, manual install — still filter for developers. A marketing lead who wants to rungtm-strategyor a compliance officer who needssoc2-readinesshits a terminal wall before seeing any output.I set up an online entry point for the library. Anyone landing on the repo can upload their project docs or deliverable drafts, describe the professional task, and get the skill's output directly:
The design decision that stood out: each skill ships with a worked example, an anti-pattern library, and a deterministic scoring rubric. That structure means the online version doesn't just run prompts — it carries the same quality guardrails into every session. A first-time user gets the rubric-scored output, not a raw LLM response.
Visitors can try any skill from the library here — upload a doc, describe the task, get the deliverable. Every run leaves a usage record on the platform, so you can see what tasks people actually bring and where the skill instructions land well or miss.
That signal loop — real inputs from non-dev users running skills on their own work — is hard to get from GitHub stars or CLI installs alone. It's the kind of feedback that sharpens SKILL.md files against actual use patterns.
shesonglin@tinkerland.ai
Feel free to close if this isn't relevant.