Agentdescent

Improve a SKILL.md, agent definition, system prompt, small codebase or host plugin by measuring it against examples and evolving it, rather than rewriting it by hand and hoping. Use whenever the user asks to improve, fix, tune, optimise, "train" or get better results out of one of those -- including when they have no test cases yet, because drafting cases for them to check is step one of the procedure, not a prerequisite for it. AgentDescent runs the edits in parallel and keeps only those that raise held-out reward.

220 stars 29 forksAI, LLMs & Agents Plugin marketplace

Install

Claude Code plugin
/plugin marketplace add ThinkFlowLab/agentdescent

The repo is a Claude Code plugin marketplace. Run /plugin afterwards to pick the plugin that includes this skill.

skills CLI (Claude Code, Cursor, Codex and more)
npx skills add ThinkFlowLab/agentdescent --skill agentdescent
Manual install
git clone --depth 1 https://github.com/ThinkFlowLab/agentdescent.git /tmp/agentdescent
mkdir -p ~/.claude/skills
cp -r /tmp/agentdescent/agentdescent/integrations ~/.claude/skills/agentdescent

Personal skills live in ~/.claude/skills. Use .claude/skills in a project folder to share a skill with your team.

From the SKILL.md

# AgentDescent

You have tools (MCP server `agentdescent`) or, without MCP, the `agentdescent`
command with the same verbs. A run is an evolution: N workers propose edits in
parallel, a merger keeps the ones that improve held-out reward, and nothing is
written back until the user says so.

**Path A only by default.** The MCP server exposes `demo` (zero-key smoke) →
doctor → plan → start → status/show → apply (plus cancel/resume, diagnose,
list/describe/algorithm_spec).
Do not reach for audit_*, release, platform, experience, or other satellites
unless the user asked and the server was started with `--advanced` /
`AGENTDESCENT_MCP_ADVANCED=1`. See docs/mcp-path-a.md.

## Zero-key try first

When the user wants to see AgentDescent work, has no cases yet, has no
provider key, or just wants to know whether a named algorithm card can run,
do this **before** The procedure below (`doctor` → `plan` → `start` need a
real model and a key):

- **With MCP:** call `demo` (optional `algorithm` such as `gepa`, `ace`,
  `dgm`, `skillopt`, `rrsi`, `archive_selection`, or `evoskill`; optional
  `kind` of `text`, `skill_dir`, or `skill_library`). It needs no key, does
  not write the user's files, and returns `initial_reward`, `final_reward`,
  `learned`, and `next`. Report the reward change and what it learned.
  `evoskill` defaults to the `skill_library` fixture.
- **Without MCP:** run `agentdescent demo [--algorithm NAME] --json` and
  report the same fields.
Read the full skill on GitHub

Details

Skill name
agentdescent
Path
agentdescent/integrations/SKILL.md
License
MIT
Repo updated
18 November 2026

FAQ

What does the Agentdescent skill do?

Improve a SKILL.md, agent definition, system prompt, small codebase or host plugin by measuring it against examples and evolving it, rather than rewriting it by hand and hoping. Use whenever the user asks to improve, fix, tune, optimise, "train" or get better results out of one of those -- including when they have no test cases yet, because drafting cases for them to check is step one of the procedure, not a prerequisite for it. AgentDescent runs the edits in parallel and keeps only those that raise held-out reward.

How do I install the Agentdescent skill?

Claude Code plugin: /plugin marketplace add ThinkFlowLab/agentdescent. Skills go in ~/.claude/skills (personal) or .claude/skills (project), and Claude loads them automatically when a task matches the description.

Is Agentdescent popular?

It lives in ThinkFlowLab/agentdescent, which has 220 GitHub stars and 29 forks.

Related skills

Fix Issue

220

ThinkFlowLab/agentdescent

Fix one AgentDescent issue end to end - claim, fresh clone, a test that fails on main, gates, PR, self-review at the tip, merge by tier, cleanup.

Testing & QA29

Review PR

220

ThinkFlowLab/agentdescent

Review an AgentDescent PR (or local change) at a tip SHA and write a Verifier report with Tip SHA / Ran / Not run / Verdict PASS|FAIL|INC that the review gate can read.

Git & Code Review29

Cavecrew

155

kubestellar/console

Decision guide for delegating to caveman-style subagents. Tells the main thread WHEN to spawn `cavecrew-investigator` (locate code), `cavecrew-builder` (1-2 file edit), or `cavecrew-reviewer` (diff review) instead of doing the work inline or using vanilla `Explore`. Subagent output is caveman-compressed so the tool-result injected back into main context is ~60% smaller — main context lasts longer across long sessions. Trigger: "delegate to subagent", "use cavecrew", "spawn investigator/builder/reviewer", "save context", "compressed agent output".

AI, LLMs & AgentsIn 1 list139

Token Budget Tracking

492

cosmicstack-labs/mercury-agent-skills

Track, optimize, and control token consumption across multi-agent systems. Covers budget allocation, real-time monitoring, cost attribution, per-agent limits, and proactive cost optimization for production LLM deployments.

AI, LLMs & Agents68

Caveman Stats

155

kubestellar/console

Show real token usage and estimated savings for the current session. Reads directly from the Claude Code session log — no AI estimation. Triggers on /caveman-stats. Output is injected by the mode-tracker hook; the model itself does not compute the numbers.

AI, LLMs & AgentsIn 1 list139

Routerbase Model Gateway

492

cosmicstack-labs/mercury-agent-skills

Configure RouterBase as an OpenAI compatible model gateway for AI apps, with routing, fallback, media generation, and credential handling patterns.

AI, LLMs & Agents68

Browse the most popular Claude skills, all ai, llms & agents skills, or write your own with the skills generator.