AgentDescent
AgentDescent is a Python framework that improves prompts, skills, and agent setups by testing them against your own examples for developers and builders.
What it is good for
- Improving a prompt or skill automatically using your own test cases
- Comparing and running named self-improvement methods on an agent
- Letting a coding agent run an evolution loop through an MCP connector
At a glance
- Who it is for
- Setup difficulty
- Developer ยท Requires programming
Level 5 of 5. You need to write code to use it.
- Cost
- Free and open source
- Type
- Framework or SDK
- Tasks
- Works with
- Claude Code, Codex, OpenCode, DeepSeek Harness, Grok Bot, Hugging Face
Momentum
- Rank this week
- Not ranked yet
- Best rank
- None yet
- GitHub stars
- 221
GitHub numbers as of Nov 16, 2026. Updated weekly.
Official links
Setup steps are on the official pages. We do not repeat them here, since they change.
Questions
- Is AgentDescent free?
- Yes. It is free and open source (MIT license).
- Do I need to know how to code to use AgentDescent?
- Yes. You need to write code to use it.
- What does AgentDescent work with?
- According to its own documentation, it works with Claude Code, Codex, OpenCode, DeepSeek Harness, Grok Bot, and Hugging Face.
- What is AgentDescent good for?
- Improving a prompt or skill automatically using your own test cases, comparing and running named self-improvement methods on an agent, and letting a coding agent run an evolution loop through an MCP connector.
Similar agents to AgentDescent
- caveman
Caveman is a skill and proxy that cuts token use by making coding agents answer tersely and shrink the text they read, for developers and heavy AI users.
- agentskills
agentskills is a specification and documentation set for Agent Skills, a standard format that gives AI agents reusable instructions for specialized tasks.
- GEPA
GEPA is a framework that improves prompts, code, and agent settings by testing and rewriting them automatically for developers.
- AssetOpsBench
AssetOpsBench is a framework and benchmark for building and testing AI agents that handle industrial equipment maintenance for researchers and engineers.
- Memanto
Memanto is a memory layer that keeps AI agents' shared knowledge current, resolving contradictions and expiring stale facts for developers running several agents.
- AgentDescent
AgentDescent is a Python framework that improves prompts, skills and agent setups against your own examples, using parallel edits and a held-out test, for developers and advanced builders.