Scene-first agent skill decisions
Find the scene before the skill.
Start with the situation a user can recognize. Each topic narrows thousands of public skills into a few practical scenes, then recommends a skill, an adaptation, a build path, or no skill.
Topic decision library
Choose from stable user scenes.
These pages are not marketplace categories. They are high-frequency decision areas where the user can recognize their situation before seeing a skill recommendation.
Design or improve a web UI
Figure out whether you need a new interface, a polish pass, a design-system pass, or critique before choosing a UI skill.
Understand, fix, or improve existing code
Separate debugging, refactoring, review, tests, and codebase understanding before choosing a coding skill.
Research a question and produce evidence-backed output
Choose between web research, literature review, market research, and document synthesis based on source type and confidence needs.
Create, revise, or extract office documents
Pick by file type and desired artifact: PDF extraction, Word editing, slides, spreadsheets, or interpretation only.
Write and repurpose content
Choose writing workflows by source material, audience, channel, and whether evidence or brand voice matters.
Improve website growth and SEO
Distinguish full-site SEO audits, on-page optimization, content planning, and AI-search visibility before choosing a growth skill.
Connect tools or automate recurring work
Choose between MCP/API building, SaaS workspace actions, recurring reports, and scheduled agent workflows.
Manage knowledge, meetings, and project context
Pick by source: meeting notes, personal knowledge, project instructions, support tickets, or team workspace memory.
Request the next decision scene.
Tell us the situation in your own words. The next page should start with the job, not a skill name.
Decision method
Topic and scene come before marketplace rank.
Popular data from skills.sh, SkillsMP, and GitHub is useful only after the user scene is stable. We do not expose language variants, install-path copies, or repository duplicates as separate user decisions.
- 01Define the topic
Use language a user would naturally say, not supply-side categories.
- 02Pick the scene
Keep only situations that actually change the recommendation.
- 03Match the skill
Recommend one primary skill, one fallback path, and visible evidence limits.