MCP Server Builder
anthropics/skills
Guides the design and implementation of Model Context Protocol servers in TypeScript or Python, from tool naming and error messages to evaluation.
Run evidence-driven evolution of agents, prompts, skills, and agent harnesses.
Apache-2.0Auto-check passed
Install Evolve Agent
$ npx skills add simple-agent-lab/RSIHub --skill evolve-agent -a claude-codeProject install by default; add -g for ~/.claude/skills/.
$ gh skill install simple-agent-lab/RSIHub evolve-agent --agent claude-codeProject scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).
$ git clone --depth 1 https://github.com/simple-agent-lab/RSIHub.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/evolve-agent .claude/skills/evolve-agent && rm -rf skills-srcUse ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.
Claude Code skills documentation · loads skills from .claude/skills/
Install the "evolve-agent" agent skill from https://github.com/simple-agent-lab/RSIHub/tree/main/skills/evolve-agent into .claude/skills/evolve-agent/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "evolve-agent", then confirm the skill loads.Claude Code copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$skill-installer install https://github.com/simple-agent-lab/RSIHub/tree/main/skills/evolve-agentType this inside Codex. $skill-installer <name> installs a curated skill from openai/skills. The installer writes to $CODEX_HOME/skills (default ~/.codex/skills). Restart Codex if the skill does not show up.
$ npx skills add simple-agent-lab/RSIHub --skill evolve-agent -a codexProject install goes to .agents/skills/; add -g for ~/.codex/skills/.
$ gh skill install simple-agent-lab/RSIHub evolve-agent --agent codexProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/simple-agent-lab/RSIHub.git skills-src && mkdir -p .agents/skills && cp -r skills-src/skills/evolve-agent .agents/skills/evolve-agent && rm -rf skills-srcUse ~/.agents/skills/ instead of .agents/skills for a personal install.
Codex skills documentation · loads skills from .agents/skills/
Install the "evolve-agent" agent skill from https://github.com/simple-agent-lab/RSIHub/tree/main/skills/evolve-agent into .agents/skills/evolve-agent/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "evolve-agent", then confirm the skill loads.Codex copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add simple-agent-lab/RSIHub --skill evolve-agent -a cursorProject install goes to .agents/skills/; add -g for ~/.cursor/skills/.
$ gh skill install simple-agent-lab/RSIHub evolve-agent --agent cursorProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/simple-agent-lab/RSIHub.git skills-src && mkdir -p .cursor/skills && cp -r skills-src/skills/evolve-agent .cursor/skills/evolve-agent && rm -rf skills-srcUse ~/.cursor/skills/ instead of .cursor/skills for a personal install.
Cursor skills documentation · loads skills from .cursor/skills/, .agents/skills/, .claude/skills/, .codex/skills/
Install the "evolve-agent" agent skill from https://github.com/simple-agent-lab/RSIHub/tree/main/skills/evolve-agent into .cursor/skills/evolve-agent/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "evolve-agent", then confirm the skill loads.Cursor copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gemini skills install https://github.com/simple-agent-lab/RSIHub.git --path skills/evolve-agent--scope user (default) or --scope workspace; --path is the subfolder of the repo that holds the skill; --consent skips the security confirmation prompt.
$ npx skills add simple-agent-lab/RSIHub --skill evolve-agent -a gemini-cliProject install goes to .agents/skills/; add -g for ~/.gemini/skills/.
$ gh skill install simple-agent-lab/RSIHub evolve-agent --agent gemini-cliProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/simple-agent-lab/RSIHub.git skills-src && mkdir -p .gemini/skills && cp -r skills-src/skills/evolve-agent .gemini/skills/evolve-agent && rm -rf skills-srcUse ~/.gemini/skills/ instead of .gemini/skills for a personal install, then run /skills reload.
Gemini CLI skills documentation · loads skills from .gemini/skills/, .agents/skills/
Install the "evolve-agent" agent skill from https://github.com/simple-agent-lab/RSIHub/tree/main/skills/evolve-agent into .gemini/skills/evolve-agent/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "evolve-agent", then confirm the skill loads.Gemini CLI copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gh skill install simple-agent-lab/RSIHub evolve-agentInstalls for Copilot at project scope by default; add --scope user for a personal install. Preview a skill first with gh skill preview. Needs GitHub CLI 2.90.0 or later (public preview).
$ npx skills add simple-agent-lab/RSIHub --skill evolve-agent -a github-copilotProject install goes to .agents/skills/; add -g for ~/.copilot/skills/.
$ git clone --depth 1 https://github.com/simple-agent-lab/RSIHub.git skills-src && mkdir -p .github/skills && cp -r skills-src/skills/evolve-agent .github/skills/evolve-agent && rm -rf skills-srcUse ~/.copilot/skills/ instead of .github/skills for a personal install. Commit .github/skills so cloud agent and code review can use it.
GitHub Copilot skills documentation · loads skills from .github/skills/, .claude/skills/, .agents/skills/
Install the "evolve-agent" agent skill from https://github.com/simple-agent-lab/RSIHub/tree/main/skills/evolve-agent into .github/skills/evolve-agent/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "evolve-agent", then confirm the skill loads.GitHub Copilot copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add simple-agent-lab/RSIHub --skill evolve-agent -a opencodeOpenCode documents no install command of its own. Project install goes to .agents/skills/; add -g for ~/.config/opencode/skills/.
$ gh skill install simple-agent-lab/RSIHub evolve-agent --agent opencodeProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/simple-agent-lab/RSIHub.git skills-src && mkdir -p .opencode/skills && cp -r skills-src/skills/evolve-agent .opencode/skills/evolve-agent && rm -rf skills-srcUse ~/.config/opencode/skills/ instead of .opencode/skills for a personal install.
OpenCode skills documentation · loads skills from .opencode/skills/, .claude/skills/, .agents/skills/
Install the "evolve-agent" agent skill from https://github.com/simple-agent-lab/RSIHub/tree/main/skills/evolve-agent into .opencode/skills/evolve-agent/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "evolve-agent", then confirm the skill loads.OpenCode copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
Facts
- Skill name
evolve-agent- GitHub stars
- 180
- Token cost
- ~2.3k tokens
- SKILL.md length
- 1,076 words
- Files
- 10 (incl. references)
- Skills in repo
- 3
- Repo updated
- First seen
- Licence
- Apache-2.0
At a glance
Run evidence-driven evolution of agents, prompts, skills, and agent harnesses.
- Works in 6 steps: Establish the contract → Choose a method and control path → Author reusable operators in a source… → …
- Asked to initialize
- SKILL.md covers 1. Establish the contract, 2. Choose a method and control…, 3. Author reusable operators… and 4. Prefer capabilities over…, plus 4 more sections
- Calls uv
What it does
Evolve Agent is an agent skill from simple-agent-lab/RSIHub. Run evidence-driven evolution of agents, prompts, skills, and agent harnesses. Use when asked to initialize or operate an evolution workspace, choose Hill Climb, A-Evolve, GEPA, AHE, or HyperAgents, let an outer Agent adapt the evolution process, invoke operators directly, improve a candidate through repeated evaluation, recover interrupted evolution, or report an evidence-backed champion.
Its SKILL.md is about 2.3k tokens, which your agent loads only when the skill is triggered. The skill folder holds 11 other files, including reference files (for example `agents/openai.yaml`, `references/a-evolve.md` and `references/agent-driven.md`).
It works with Python. The repository describes itself as: A research framework for principled agent self-improvement under frozen evaluators and declared mutation boundaries, recording verifiable lineage to make it reproducible and… The licence is Apache-2.0.
When your agent uses it
- Asked to initialize
- Operate an evolution workspace
- Choose Hill Climb
- Let an outer Agent adapt the evolution process
Example prompts
- “/evolve-agent”
Workflow steps
6 steps, taken from the step headings in SKILL.md.
What it can do on your machine
Read from SKILL.md and the folder at commit 64491da. It shows what the files ask for, not the result of running them.
Tool permissions
Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.
From allowed-tools in the SKILL.md frontmatter.
Runs code
Shell commands in SKILL.md call:
uv
From the folder's file list and the shell code blocks in SKILL.md.
Network
No URLs in SKILL.md. Its commands use uv, which can reach the network depending on how they are called.
From URLs in SKILL.md, links to its own repository left out.
Credentials
Names no API keys, tokens, secrets or passwords.
From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.
Context cost
Evolve Agent loads about 2.3k tokens when it runs, and up to ~11k if it reads all its reference files. Until then it costs about 101 tokens; SKILL.md has 1,076 words of instructions outside code blocks.
Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.
Safety
Auto-check passed
The automated check found no risky patterns in SKILL.md.
Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.
SKILL.md
The full file from simple-agent-lab/RSIHub at commit 64491da, republished under its Apache-2.0 licence (© simple-agent-lab). 1,076 words, ~2,302 tokens.
Download SKILL.mdSave it as
.claude/skills/evolve-agent/SKILL.md (or your agent's skills folder). This skill also uses 9 other files; get the full folder from GitHub.- name
- evolve-agent
- description
- Run evidence-driven evolution of agents, prompts, skills, and agent harnesses. Use when asked to initialize or operate an evolution workspace, choose Hill Climb, A-Evolve, GEPA, AHE, or HyperAgents, let an outer Agent adapt the evolution process, invoke operators directly, improve a candidate through repeated evaluation, recover interrupted evolution, or report an evidence-backed champion.
Build an evidence chain
Treat evolution as an evidence chain:
contract → baseline → evidence → hypothesis → candidate → evaluation → lineageA higher score alone is insufficient. Link every candidate to the evidence that motivated it, its exact snapshot, frozen evaluation, and lineage decision.
1. Establish the contract
Detect whether the current directory is an initialized evolution workspace.
- In a workspace, read
AGENTS.md,evolve.yaml,program.md, then run./evolve status .and./evolve verify .. - For a new experiment, identify the target, mutable surface, evaluator, data partitions, budget, and execution boundary before initialization.
- Read the workspace contract before creating, operating, recovering, or interpreting a workspace. Its "Create a workspace" section gives the initialization and baseline-certification commands and the preconditions they enforce.
Completion check: Name the target, mutable surface, frozen evaluator (the
evaluator/ contract that scores every candidate), data
partitions, candidate budget, and execution boundary. In an existing workspace,
also identify the current champion, next generation, and interrupted state.
2. Choose a method and control path
The initialized operators define the starting method. The control path may be
driver-led, where evolve run fixes the lifecycle, or agent-led, where the
outer Agent chooses which direct capabilities to invoke and may change the
active process when the surface permits it. For a new experiment, GEPA is the
default; choose Hill Climb when the experiment needs the simplest attributable
control. Match the method and control path to the research question, available
evidence, and mutable surface. Read only the relevant method card.
| Observable condition | Method | Read |
|---|---|---|
| A minimal attributable control is enough | Hill Climb | hill-climb.md |
| Behavioral traces or generated-artifact rubrics should guide prompt or skill mutation | A-Evolve | a-evolve.md |
| The evaluator returns per-task results and the target splits into components | GEPA | gepa.md |
| Failures are execution-shaped and justify harness changes | AHE | ahe.md |
| The evolution process itself may also change | HyperAgents | hyperagents.md |
When the outer Agent, rather than a configured mutate stage, should decide the sequence of investigation, operator calls, edits, retries, and stopping, read agent-driven control. This is an experimental control path over existing workspace capabilities, not a new operator method.
Read scientific foundations only when defining or changing evaluator semantics, partitions, acceptance rules, or research claims.
Completion check: State why the method and control path match the evidence and declared mutable surface. If they do not, choose again before running.
For artifact-producing Skills, prefer replaying a selected parent's certified artifacts over executing that parent again. Re-execute the parent only when the current task set, evaluator identity, runtime identity, or required artifacts do not match the retained evidence. Execute every child freshly.
3. Author reusable operators in a source checkout
Use the library when creating a reusable policy, not an edit to one already initialized workspace. Discover available entries, then scaffold and verify one operator before selecting it from a recipe:
uv run --frozen evolve operator list [stage]
uv run --frozen evolve operator new mutate <name>
uv run --frozen evolve operator describe mutate/<name>
uv run --frozen evolve operator check mutate/<name> --config '{"attempts": 3}'
uv run --frozen evolve recipe check <recipe-path>new writes exactly one entry at library/mutate/<name>.py. Implement the
generated MutateOperator, keep validate_config, and use
sdk.main(..., validate_config=validate_config). A recipe selects it with an
operator: value and nested config: mapping:
operators:
mutate:
operator: critic_editor
timeout_s: 3600
config:
attempts: 3Do not put a reusable implementation beside a recipe or alter a library entry to change a running workspace. Run recipe check before initialization; a new workspace freezes the selected source. Existing workspaces retain their own frozen active operators.
Completion check: The operator is in the central library, its configuration
passes operator check, the recipe passes recipe check, and the source change
is separated from any initialized workspace it does not retroactively alter.
Show full SKILL.md (524 more words)Show less
4. Prefer capabilities over source
For agent-led evolution, start from the stable workspace interface:
./evolve operator active . --json
./evolve operator run . <stage> --genid <id> [stage arguments]Treat operator active --json as the live authority for which stages are
configured and whether their access is direct, driver, or finalize.
Invoke configured direct operators, read their retained artifacts under
runs/gen-<id>/, and make the candidate change yourself.
Escalate progressively:
- Tune one call with
--configwhen the capability is right but its bounds are wrong. - Read
PROTOCOL.md, operator guidance, oroperators/README.mdwhen an input or artifact is unclear. - Read the active
operators/<stage>.pyonly to diagnose behavior or change the active evolution process. - Read
library/<stage>/only to compare or adapt another implementation.
Do not read implementation source merely to invoke a working operator. Do not
edit library/ and assume runtime behavior changed; active code lives under
operators/.
Use the configured driver when its mutation stage should own the edit and an unattended run is desired:
./evolve run . --max-generations 1Driver and agent-led paths share the same evaluation and lineage mechanism. Do not run them concurrently. Ordinary agent-led work should close one generation through the stable commands below. For an explicitly Agent Driven experiment, the outer Agent may adapt its action sequence under the Agent Driven control reference; it must still use the mechanism for candidate identity, evaluation, and finalization.
Completion check: Choose exactly one control path for the active work. For agent-led evolution, name the available direct operators, hard budget, and the evidence supporting the next action; source inspection must have a concrete reason.
5. Close the loop
- Establish and inspect the certified baseline.
- Select a parent and retain the method's required evidence.
- State one evidence-linked hypothesis and predicted effect.
- Produce one candidate inside the declared mutable surface.
- Run every configured admission check against the final candidate snapshot.
- Evaluate and finalize through the workspace mechanism.
- Verify lineage before beginning another generation.
Completion check: The candidate has an exact lineage identity; required admission decisions and evaluator-stamped results exist; lineage verification passes; accepted and rejected outcomes remain auditable.
6. Report only what the chain proves
Start from ./evolve report ., which writes the experiment report and
research-claim checklist from stamped records. Around it, report the baseline,
champion, parent-child changes, accepted and rejected mutations, evaluation
scope, retained evidence, and limitations. Tie every quality claim to
evaluator-stamped artifacts from the run.
Completion check: Every score and champion identity is derivable from trusted lineage records, and every generalization claim names its data partition.
Guard the chain
- Keep one evaluator and runtime identity within an experiment. Start a new experiment when the evaluator changes.
- Take scores and champion state only from mechanism-owned stamped records.
- Keep optimization, gate, and sealed task identities disjoint.
- Change only the declared mutable surface.
- Treat linked worktrees outside
runs/worktrees/as user-owned. Report them; never remove or modify them without explicit authorization. - Match the execution boundary to candidate trust.
- Keep credentials out of prompts, artifacts, and reports.
- Spend live evaluation budget only when the request authorizes execution.
Historical-workspace note
Older initialized workspaces retain the stage files and configuration frozen at creation time. Treat those as historical metadata only; start a new workspace to use the current operator model.
© simple-agent-lab, Apache-2.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file
Files
SKILL.md and 9 other files (references) in skills/evolve-agent of simple-agent-lab/RSIHub.
- SKILL.md
- agents/openai.yaml
- references/a-evolve.md
- references/agent-driven.md
- references/ahe.md
- references/gepa.md
- references/hill-climb.md
- references/hyperagents.md
- references/scientific-foundations.md
- references/workspace-contract.md
Open the folder on GitHubat commit 64491da
Compare with similar skills
Evolve Agent next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.
| Skill | Stars | Used in | Tokens | Auto-check | Licence | Repo updated |
|---|---|---|---|---|---|---|
| Evolve Agent this skillsimple-agent-lab/RSIHub | 180 | — | ~2.3k | Automated safety check: Pass | Apache-2.0 | |
| MCP Server Builderanthropics/skills | 180k | 64 repos | ~2.3k | Automated safety check: Pass | Apache-2.0 | |
| PDF Processinganthropics/skills | 180k | 48 repos | ~2k | Automated safety check: Pass | Proprietary | |
| NotebookLM Research AssistantPleasePrompto/notebooklm-skill | 7.8k | 14 repos | ~2.4k | Automated safety check: Notes | MIT | |
| Manim Video Productionbrowser-use/video-use | 28k | 6 repos | ~3k | Automated safety check: Pass | MIT | |
| Code Review ChecklistshareAI-lab/learn-claude-code | 78k | 5 repos | ~1.1k | Automated safety check: Pass | MIT |
Similar skills
- Official
- Official
PDF Processing
anthropics/skills
Handles everyday PDF jobs in Python and on the command line: extract text and tables, merge, split, rotate, watermark, fill forms, encrypt and OCR.
NotebookLM Research Assistant
PleasePrompto/notebooklm-skill
Lets Claude Code ask questions of your Google NotebookLM notebooks through browser automation and return answers grounded in your uploaded sources.
Manim Video Production
browser-use/video-use
Produces math and technical explainer videos with Manim Community Edition: concept animations, equation derivations, algorithm walkthroughs and data stories.
Code Review Checklist
shareAI-lab/learn-claude-code
Reviews code against a five-part checklist covering security, correctness, performance, maintainability and testing, and reports findings in a fixed format.
PPT Master
hugohe3/ppt-master
Generates editable PowerPoint decks, rebuilds slides from images, fills .pptx templates and polishes existing presentations through routed workflows.
More from simple-agent-lab/RSIHub
Task Execution
simple-agent-lab/RSIHub
Solve repository tasks by inspecting first, making focused edits, and verifying the result.
Task Execution
simple-agent-lab/RSIHub
Baseline procedure for completing a terminal task reliably. An agent skill from simple-agent-lab/RSIHub.
Related
Works with
Questions about Evolve Agent
What does Evolve Agent do?
Run evidence-driven evolution of agents, prompts, skills, and agent harnesses. Evolve Agent is an agent skill from simple-agent-lab/RSIHub. Run evidence-driven evolution of agents, prompts, skills, and agent harnesses.
When should I use Evolve Agent?
Evolve Agent fits situations like: asked to initialize; operate an evolution workspace; choose Hill Climb; let an outer Agent adapt the evolution process.
How do I install Evolve Agent in Claude Code?
Run `npx skills add simple-agent-lab/RSIHub --skill evolve-agent -a claude-code`. Or copy the skill folder (skills/evolve-agent in simple-agent-lab/RSIHub) into .claude/skills/evolve-agent in your project. Claude Code loads it when a task matches its description.
How do I install Evolve Agent in Codex?
Run `npx skills add simple-agent-lab/RSIHub --skill evolve-agent -a codex`. Or copy the skill folder (skills/evolve-agent in simple-agent-lab/RSIHub) into .agents/skills/evolve-agent in your project. Codex loads it when a task matches its description.
Can I use Evolve Agent in Cursor, Gemini CLI or GitHub Copilot?
Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add simple-agent-lab/RSIHub --skill evolve-agent -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/evolve-agent, .gemini/skills/evolve-agent, .github/skills/evolve-agent and .opencode/skills/evolve-agent in your project.
What does Evolve Agent need to run?
Going by SKILL.md and its folder, Evolve Agent needs the command-line tools its instructions call (uv).
Does Evolve Agent access the network?
SKILL.md contains no URLs. Its commands use uv, which can reach the network depending on how they are called. This is read from the text; nothing was executed.
Is Evolve Agent safe to install?
Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.
What licence does Evolve Agent use?
Evolve Agent is published under the Apache-2.0 licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.
How many tokens does Evolve Agent use?
About 2.3k tokens (SKILL.md is roughly 9.2k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 9.2k tokens, read only when the agent opens those files.
What are the alternatives to Evolve Agent?
Skills that share tags, products or a category with Evolve Agent: MCP Server Builder (anthropics/skills, 180k stars), PDF Processing (anthropics/skills, 180k stars), NotebookLM Research Assistant (PleasePrompto/notebooklm-skill, 7.8k stars) and Manim Video Production (browser-use/video-use, 28k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.
Who maintains Evolve Agent?
simple-agent-lab (a GitHub organization) maintains it in simple-agent-lab/RSIHub, which has 180 GitHub stars. The repository holds 3 skills in this directory. The repository was last updated on September 13, 2026.
Source: simple-agent-lab/RSIHub on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.