← Open Skills overview
Open Skills directory.
Do not browse this like an app store. Start with the repeated failure: the agent keeps forgetting a tool shape, writing in the wrong voice, losing verification knowledge, mishandling media, or needing the same setup speech every session.
Then choose the smallest primitive that fixes that failure. Each entry tells you what the skill does, why it is worth building, what it depends on, and gives your agent a setup prompt to adapt the primitive to your stack and taste.
The useful skill is the one that removes a repeated explanation.
41 skills8 categories
Choose by bottleneck, not by novelty.
If you cannot name the workflow where a skill will run next week, wait. Open Skills works because each primitive is attached to a real recurring job.
Core Infrastructure5 skillsThe foundation layer: image generation, current search, transcription, ingestion, and artifacts. Build these when other workflows keep re-solving the same tool and packaging problems.Context Engineering10 skillsSkills for controlling what actually reaches the model: keep the context budget honest, then turn scattered, uncited paperwork into a structured case file — ingest documents, normalize records, store evidence, retrieve deterministically, validate citations, and export a human-reviewed packet.Research & Thinking5 skillsSkills for turning messy input into thinking you can review: voice notes, meetings, document piles, weekly noise, and assumptions that need to be challenged before they become plans.Writing, Voice & Content4 skillsSkills that make agent writing specific: a real voice, a real audience, current facts, branded images, and publishing formats that keep content from sliding back into generic AI prose.Web Publishing & Frontend4 skillsSkills for turning agent output into public, inspectable web work: better frontend taste, clean publishing, share previews, comparison pages, and verification before anything is called shipped.Video & Media Production3 skillsSkills for the expensive parts of media work: transcript-first editing, motion graphics, timeline assembly, and NLE control. Build these after the simpler media primitives are working.Testing & Quality3 skillsSkills that make agent-built work trustworthy: repeatable QA, browser evidence, repo-local testing memory, and the habit of leaving verification knowledge where the next session can use it.Agent Operations7 skillsMeta-skills for running agents without becoming the bottleneck: goal prompts, visible delegation, operating maps, merge discipline, stakeholder updates, and skill-library compounding.