AI Agents Catalog

Claude Agents are autonomous AI programs that plan, decide, and act across multi-step tasks - using tools, reading context, evaluating results, and iterating until the job is done. Unlike a skill or prompt that responds to a single instruction, an agent runs a loop.

Agents vs Skills vs Prompts

A prompt is a single instruction. A skill is a reusable instruction module. An agent is a program: it reads context, selects tools (MCP servers, code runners, APIs), evaluates output, branches on results, and continues until completion. Agents handle tasks too complex for a single-step response - refactoring a codebase, researching a market, triaging an inbox.

Agent formats on Spark

The most common format is a CLAUDE.md or project instruction file: a context block loaded at the start of every Claude Code session that shapes the agent's role, tools, and objectives for a specific project type. Other formats include agent definition files for frameworks like AutoGen, CrewAI, and LangGraph; and full agent repositories with a running backend loop. Many agents on Spark ship with companion skills and MCP connectors as a ready-to-install stack.

Use cases by job cluster

Coding agents handle code review, PR triage, test generation, refactoring, and documentation. Research agents run literature search, competitive analysis, and market mapping. Operations agents handle scheduling, email triage, and CRM updates. Creative agents manage content drafting, SEO writing, and social scheduling.

What to look for

Agents above score 75 have been independently tested, have active upstream maintenance, and include clear setup instructions. Filter by integration tag to find agents already wired for the tools you use - GitHub, Notion, Slack, or specific cloud providers.