ahc-optuna
A skill for safely tuning AHC hyperparameters using Optuna and validating results.
Install
mkdir -p .claude/skills/ahc-optuna && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/12762" && unzip -o skill.zip -d .claude/skills/ahc-optuna && rm skill.zipInstalls to .claude/skills/ahc-optuna
Activation
This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.
AHC のハイパーパラメータ最適化(`eval.py -o`)を安全に回すための skill。Key capabilities
- →Optimize hyperparameters for AHC
- →Update default values in a.rs
- →Verify score differences with extreme values
- →Report changed parameters and re-validation results
How it works
This skill automates hyperparameter optimization for AHC by running evaluations, updating default values in a.rs, and verifying improvements.
Inputs & outputs
When to use ahc-optuna
- →Optimizing hyperparameter constants for AHC
- →Running automated parameter evaluation sweeps
- →Updating default configurations after optimization
- →Verifying parameter changes against test sequences
About this skill
ahc-optuna
AHC のハイパーパラメータ最適化(eval.py -o)を安全に回すための skill。
使うタイミング
- ユーザーが Optuna で定数調整を依頼したとき。
- 環境変数経由で
a.rsの評価パラメータを最適化したいとき。
MUST
- 対象定数は
*_DEFAULTとしてa.rs冒頭に定義する。 Envに対応メンバを追加し、Env::initで初期化する。- 初期化は次の形式にそろえる:
self.xxx = os_env::get("xxx").and_then(|s: String| s.parse::<f64>().ok()).unwrap_or(XXX_DEFAULT);
eval.py側はAHC_PARAMS_XXX、a.rs側はos_env::get("xxx")を使う。- 最適化対象は実際の選択ロジックで使う値のみに限定する。
- 事前チェック:
/Users/toastuz/Develop/.venv/bin/python3 eval.py -s 0 -v- 極端値でスコア差分を確認(キー反映漏れ検知)
- 実行:
- 短試行:
/Users/toastuz/Develop/.venv/bin/python3 eval.py -s 0 49 -o 20 - 本試行:
/Users/toastuz/Develop/.venv/bin/python3 eval.py -s 0 49 -o 100
- 短試行:
- 反映:
study.best_trial.paramsを*_DEFAULTへ反映する。/Users/toastuz/Develop/.venv/bin/python3 eval.py -s 0 -v/Users/toastuz/Develop/.venv/bin/python3 eval.py -s 0 49- 必要なら
/Users/toastuz/Develop/.venv/bin/python3 eval.py -s 0 49 --seq
SHOULD
- 探索範囲は現行値の近傍から開始し、端に寄ったら再設定する。
- 一度に最適化するパラメータ数は抑える。
- 実行結果は「変更パラメータ」「best」「再検証結果」をセットで報告する。
FORBIDDEN
a.rs側でAHC_PARAMS_接頭辞付きキーを直接参照しない。- 未使用パラメータを最適化対象に入れない。
- 全試行同値を放置して次ステップへ進めない。
OUTPUT
- 変更した
*_DEFAULT一覧 - best trial の主要パラメータ
- 再検証結果(
-s 0 -v,-s 0 49, 必要なら--seq)
When not to use it
- →When a.rs directly references AHC_PARAMS_ prefixed keys
- →When unused parameters are included in optimization
- →When all trials yield the same value
Limitations
- →It is specific to AHC hyperparameter optimization
- →It requires parameters to be defined as *_DEFAULT in a.rs
- →It requires specific environment variable handling for optimization
How it compares
This skill provides a structured and safe workflow for hyperparameter optimization, including specific file modification and validation steps, unlike a generic optimization process.
Compared to similar skills
ahc-optuna side by side with the closest alternatives in the catalog.
| Skill | Installs | Updated | Safety | Difficulty |
|---|---|---|---|---|
| ahc-optuna (this skill) | 0 | 5mo | No flags | Intermediate |
| rfx-bootstrap-install | 0 | 6mo | Review | Beginner |
| tui | 0 | 4mo | No flags | Intermediate |
| pdf-processing-pro | 17 | 10mo | Review | Intermediate |
Try saying
Example prompts that trigger this skill in your AI assistant.
You might also like
rfx-bootstrap-install
quantbagel
Bootstrap the rfx monorepo for local development. Use when asked to install dependencies, enable hooks, or prepare this repo for an agent to work safely.
tui
re-cinq
Expert terminal user interface development including interactive console applications, cross-platform TUI libraries, and responsive terminal layouts
pdf-processing-pro
davila7
Production-ready PDF processing with forms, tables, OCR, validation, and batch operations. Use when working with complex PDF workflows in production environments, processing large volumes of PDFs, or requiring robust error handling and validation.
python-repl
gptme
Interactive Python REPL automation with common helpers and best practices
polars
davila7
Fast DataFrame library (Apache Arrow). Select, filter, group_by, joins, lazy evaluation, CSV/Parquet I/O, expression API, for high-performance data analysis workflows.
math-router
parcadei
Deterministic router for math cognitive stack - maps user intent to exact CLI commands