[19.4.0] - 2026-10-11 - "Prompt Handoffs And Cost-Gated Scaling"
Adds 2 reviewed skills for turning a rough request into a clear prompt for another agent and for proving a data/AI workflow end to end on a tiny sample before scaling; it also refreshes the Datacircle LinkedIn lookup skill with the Fetchin provider, corrects the OrcaPromptVault artifact wording, and ships guided discovery, collections and a Core overview on the hosted catalog. Ships 2,677 skills.
A catalog release for Claude Code, Cursor, Codex CLI, Gemini CLI, and related AI coding assistants. It includes the protected maintainer batch merged after 19.3.0, with existing installation interfaces preserved.
Start here
- Install:
npx agentic-awesome-skills@19.4.0 - Choose your tool
- Best skills by tool
- Bundles
- Workflows
Added
- kernel-prompt - turn a rough request or draft into a clear, self-contained prompt for another agent: read the context it mentions, ask one round of decisions only the user can make, then return only the prompt and never perform the task it describes.
risk: safe, MIT, community source codegiveness/kernel-prompt. - skeleton-first - prove a data/AI/agent workflow end to end on the smallest input with the cheapest settings, look for cheap wins in the sample (duplicates, boilerplate, invalid inputs that are still billed), then scale in cost-gated rungs within a stated budget and keep a run log of cost and decisions.
risk: safe, self-authored source.
Changed
- datacircle - add Fetchin as the third LinkedIn profile provider (price, limit, request, response and
404), use HarvestAPI or Fetchin past Up2Data's daily limit, correct Up2Data's per-profile price, and drop the retiredget_invite_linkMCP tool from the skill.risk: criticalis preserved. - glasser - add a source parameter to the maintained setup-instruction link so the provider can see which listings agents follow; the link target and installation-approval guidance are unchanged.
risk: criticalis preserved. - system-prompt-lookup - correct the second artifact label to "captured off the wire or reported by the model itself", so a model's account of its own prompt is not presented as vendor-published material.
risk: safeis preserved.
Site and catalog
- Guided discovery, skill collections and a Core overview on the hosted catalog, plus shared-shortlist review for team adoption and clearer information pages.
Credits
- @codegiveness for
kernel-prompt - @danhe6 for
skeleton-first - @waynehamadi for the Datacircle Fetchin provider update
- @glasserai for the Glasser setup-instruction link
- @xizhuomengcontin for the OrcaPromptVault wording correction
- @markqian for the openCanviz official source