TalentDeltaDownload Agents / Find Humans
We test, review and share agent templates — creating a safe space for human experts to collaborate and promote their agent designs.
Verify your email, then download agents for:
New here? Read the quickstart guide.
- Claude Code
- Claude Cowork
- Claude Desktop
- Codex CLI
- Cline
- Gemini CLI
- Hermes Agent
- OpenClaw
- OpenCode
- OpenHands
- PicoClaw
- ZeroClaw
Filtered by skill: Multi-Agent Coordination — clear
11 templates published
Building-design agent team: an architect-lead directs 12 AEC specialists — drawing producer and reviewer, BIM/IFC automation, quantity audit, life-safety, building-performance and site-code review, construction admin — over 30 skills. A brief becomes concept drawings, layered DXF sheets, schedules and revision-impact reports. Runs in Claude Code, Codex or Gemini CLI; Python 3 drives the drawing engine. Conceptual output, not statutory approval. MIT plus a GPL-2.0-or-later Blender worker.
Uploaded by TalentDelta Team Member (Scout) · via TalentDelta
Self-hosted YouTube channel operator: seven agents research topics, write scripts, design thumbnails, optimise SEO, render real MP4s from narration and visuals, schedule approved uploads, then fold retention and comment evidence into the next cycle. Human approval gates every publish. Node 18.17+, Express dashboard, SQLite, FFmpeg, 8 text providers and 5 AI video providers or free local slideshow rendering.
Uploaded by TalentDelta Team Member (Scout) · via TalentDelta
Firm-style multi-agent trading desk: four analysts (market, fundamentals, news, sentiment) run in parallel, bull and bear researchers debate them, a trader drafts the plan, three risk debaters argue it and a portfolio manager approves or rejects. Twelve agent roles with their own system prompts and twelve data tools (prices, indicators, filings, news, macro) wired as a LangGraph, plus CLI, decision log, checkpoint resume, backtests and Docker. Python 3.11+, research only: it places no orders.
Uploaded by TalentDelta Team Member (Scout) · via TalentDelta
An orchestrated development team of 33 Claude Code subagents: a team-configuration entry layer, core specialists (code reviewer, code archaeologist, performance optimiser), orchestrators, and full-stack, frontend, backend, mobile, DevOps, QA, security and data specialists, each with role, deliverables and hand-off contracts. Includes the CLAUDE.md team wiring, guides on creating agents and their dependency map, and the MIT LICENSE.
Uploaded by TalentDelta Team Member (Scout) · via TalentDelta
Code ReviewCode Refactoring and OptimizationRoot-Cause DebuggingUnit and Integration TestingTest Strategy and TDDProject DocumentationGit WorkflowPull Request ManagementFrontend DevelopmentBackend DevelopmentFull-Stack DevelopmentSystem DesignAPI DesignMulti-Agent CoordinationRole AssignmentTask DecompositionSoftware EngineeringAI AgentsLarge Language ModelsProductivity and CollaborationAutomation
Why collaborate?
- Automation leads to repeatable improvements
- Small tweaks can lead to huge gains
- Would your agent benefit from human ingenuity and professional insight?
The case for Humans
- AI agents are incredible for efficiency and consistency, but lack human ingenuity to optimise
- AI agents can search for humans with lived professional experience to help optimise them.
- One small tweak by a professional can yield massive improvements.
Collaborate
Collaborate on agent development
Download an agent template or create one from scratch
Test and improve the agent's setup
Export it from your agent harness
Upload it privately and invite users to collaborate with you
If you want to give back to the community, release the updated agent to the public.
All agent uploads are vetted and tested by TalentDelta before they can be accessed on our site, promoting safe collaboration.
Every agent template is credited to the human who uploaded it.
How a template gets here
Four steps, and every one of them shows on the template's own page.
You upload it
One line about what it does. Until it publishes, only you and the people you name can see it.
A reviewer reads it
If it is turned down, the uploader gets the reason by email.
It publishes with its history
Your name goes on it. Every build after that says what changed.
Someone downloads it
They run it in their own tooling. That count is what ranks this page.
New here? Open any template, read its README, then run the quickstart.