agentsclimarketplace

Backtest integrity

Skill alidevai/chiron_agent/.claude/skills/backtest-integrity

Autonomous, self-improving AI agent platform for Claude Code — safe capability & skill acquisition, MCP/tool discovery, sandboxed policy gates, and token-efficient learning.

Install
npx -y skills add alidevai/chiron_agent --skill backtest-integrity

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

3 things to look at

  • 20 days oldThe repository was created 20 days ago. New is not bad, but a brand new repository carrying a familiar-sounding name is the shape a typosquat arrives in, and there has been no time for anyone else to find a problem with it.
  • no licenseNo license file was found in the repository. Code published without one is not open source by default, so using it at work is a question for whoever answers licensing questions where you are.
  • 1 stars1 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

What its author says it does

Copied from the file, not written here

Bir trading backtest'inin sonuclarini guvenilir kabul etmeden once butunluk denetimidir. Look-ahead (zaman sizintisi), survivorship bias, maliyet/slippage eksigi ve asiri iyimser yorumu yakalar. Herhangi bir strateji/indikator backtest edildiginde, canli veya paper trade DUSUNULMEDEN once zorunludur.

SKILL.md

2.6 KB, as published. Nobody here has run it

Backtest Integrity

Bir backtest'in yuksek getiri gostermesi onu dogru yapmaz. Bu skill, sonucu "guvenilir" ilan etmeden once gecmesi gereken kapilari tanimlar.

Ne zaman kullanilir

Herhangi bir strateji, indikator veya sinyal backtest edildiginde. Paper trade veya canli isleme gecis DUSUNULMEDEN once. Bu bir insan-onayi kapisidir; agent tek basina canli isleme geciremez (policies/trading.yaml).

Zorunlu kontrol listesi

  1. Look-ahead bias (zaman sizintisi): Her hesaplama yalnizca o an KAPANMIS bar'lara mi dayaniyor? Gelecek bilgisi (kapanmamis bar, ileriki fiyat, yeniden ifade edilen veri) sinyale karisiyor mu? Repaint eden indikatorleri isaretle.

  2. Warm-up dogru mu: Indikatorun isinma periyodu once atlaniyor mu, yoksa eksik veriyle uretilen ilk sinyaller islenmis mi?

  3. Survivorship bias: Evren yalnizca hayatta kalan enstrumanlardan mi olusuyor? Delist olanlar, iflaslar hesaba katilmis mi?

  4. Transaction cost + slippage: Komisyon, spread ve slippage modellenmis mi? Sifir maliyetli backtest gecersizdir.

  5. Out-of-sample + walk-forward: Parametreler tek bir donemde mi optimize edildi? Gorulmemis veride (OOS) ve ileri-donuslu (walk-forward) test edildi mi?

  6. Overfitting: Kac parametre denendi? Parametre hassasiyeti (kucuk degisiklikte sonuc cokuyor mu)? Cok sayida denemeden en iyisini secmek (p-hacking) var mi?

Kalite kapilari

  • Yukaridaki kontrollerden herhangi biri BASARISIZ ise sonuc "guvenilir degil".
  • Golden dataset varsa referans hesaplama ile birebir karsilastir.
  • Sonucu evaluator subagent'ina bagimsiz dogrulat (isi yapan onaylamaz).

Trading guvenlik hatti (policies/trading.yaml)

arastirma -> backtest -> BAGIMSIZ AUDIT -> OOS -> walk-forward
  -> maliyet/slippage -> paper trade (insan onayi) -> risk limitleri
  -> INSAN ONAYI -> kucuk sermaye -> canli izleme + kill switch
  • withdrawal, limitsiz trade, agent'in kendi risk limitini degistirmesi: yasak.
  • paper_trade ve live_trade: insan onayi zorunlu.

Cikti formati

Her kontrol icin PASS/FAIL ve gerekce ver. Genel karar: RELIABLE / NOT_RELIABLE / INSUFFICIENT_EVIDENCE. Asla "strateji karli" deme; "denetimden gecti/gecmedi" de.

Kaynaklar

  • Referans araclar: Qlib, Freqtrade, CCXT (skill degil, tool/framework).
  • Ilgili: [[capability-gap-analysis]]

Keep looking

Skills are one crate of 328,083. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.