Getting it into your agent
It runs from inside its repository, so the clone comes first — what it calls does not travel with the file alone.
git clone --depth 1 https://github.com/ckorhonen/claude-skillsnpx agentmods add skills/ckorhonen/claude-skills/hyperagentWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/ckorhonen/claude-skills/hyperagent)<a href="https://agentmods.dev/skills/ckorhonen/claude-skills/hyperagent"><img src="https://agentmods.dev/badge/skills/ckorhonen/claude-skills/hyperagent.svg" alt="Measured on agentmods" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00084 | $0.03127 |
| Opus 5 | $0.00042 | $0.01563 |
| Sonnet 5 | $0.00017 | $0.00625 |
| Haiku 4.5 | $0.00008 | $0.00313 |
Grade A, and why
hyperagent scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 8d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 369 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Hyperagent
Quick Start — Simple Examples
New to Hyperagent? Try these beginner-friendly tasks before the full setup.
1. Optimize a simple Python script to run faster
Say: "Use hyperagent to optimize this script for speed" and paste something like:
# slow_sort.py
def sort_numbers(nums):
result = []
while nums:
smallest = min(nums)
result.append(smallest)
nums.remove(smallest)
return result
Hyperagent will benchmark it, propose a faster implementation, and validate the improvement.
2. Improve a prompt to get better answers
Say: "Run hyperagent on this prompt and improve accuracy" with a prompt like:
Summarize this article in one sentence.
The meta-agent iterates on the prompt, measures quality, and keeps improvements that score higher.
3. Make a sorting function more efficient
Say: "Evolve this function with hyperagent" and paste any function. Hyperagent creates a benchmark, runs generations of improvements, and shows you the performance gain per generation.
4. Self-improve any script
Say: "Self-improve this agent/script" and point to any Python file. Hyperagent wraps it in an evaluation loop, proposes modifications, and tracks what works.
The simplest possible setup: create
task.shthat printsMETRIC score=0.5, then runpython3 scripts/init_session.py. From there the loop is fully automated.
Self-referential self-improvement: a meta-agent that modifies a task-agent (and itself) to optimize any measurable objective.
Inspired by Facebook Research's Hyperagents paper (arXiv:2603.19461), which demonstrated that agents combining a task-solver and a self-modifying meta-level into a single editable program can achieve open-ended, compounding improvements that transfer across domains.
How It Works
A hyperagent is a system with two components in a single editable codebase:
- Task Agent — solves the target task (benchmark, code generation, data processing, etc.)
- Meta Agent — analyzes task performance history and proposes modifications to the task agent's code (and optionally its own code)
What ships with it
14 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
- DOCS.md 12 KB
- scripts/common.py 23 KB runs code
- scripts/init_session.py 5.6 KB runs code
- scripts/log_variant.py 4.4 KB runs code
- scripts/render_report.py 1.2 KB runs code
- scripts/run_task.py 6.2 KB runs code
- scripts/select_parent.py 2.3 KB runs code
- tests/checks.sh 281 B runs code
- tests/DOCS.md 2.4 KB
- tests/target_v1.py 1.6 KB runs code
- tests/target_v2.py 1.8 KB runs code
- tests/target_v3.py 2.2 KB runs code
- tests/target.py 1.5 KB runs code
- tests/task.sh 138 B runs code
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 8d ago First seen · 369 lines · 84 tokens per session scan A b882ba44a410
hyperagent is a skill published in the GitHub repository ckorhonen/claude-skills (14 stars, last pushed 2mo ago), licensed MIT. It adds 84 tokens to every session and 3,127 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
watch
File sentinel that monitors the working directory for changes and marker comments, then auto-triggers appropriate skills. Poll-based via git diff against the last scan commit. Writes intake items for batch processing and routes marker actions through /do. Use for automatic reactions to file changes; do NOT use for…
review
5-pass structured code review — correctness, security, performance, readability, consistency.
live-preview
Mid-build visual verification loop. Takes screenshots of components during construction, not just after. Catches visual regressions and invisible features before they compound. Requires Playwright or similar screenshot tool.
wiki
Markdown-first knowledge base where the LLM acts as librarian. Ingests raw sources, compiles and interlinks topic files, self-maintains an index. No vector DB or embeddings required -- uses LLM-native navigation over structured markdown up to 400K words.
huggingface-llm-trainer
Train or fine-tune language models with TRL or Unsloth on Hugging Face Jobs, including SFT, DPO, GRPO, reward models, and GGUF conversion. Use for cloud LLM training; use huggingface-vision-trainer for vision tasks.
feature-dev
Guide a feature implementation through a structured seven-phase workflow with deep codebase understanding, clarifying questions, parallel architecture design, and quality review. Use this skill when the user asks to build a new feature, add functionality, or wants a methodical approach to implementation rather than…