ワンクリックでManusで任意のスキルを実行

context-budget

スター561

フォーク103

更新日2026年6月29日 22:25

Keep the agent's context lean so accuracy doesn't collapse. Use on long sessions, big files, or when the agent starts hallucinating.

インストール

Codex または Claude でインストールこの Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。

Manusで実行

ソース

Archive228

Archive228/loopkit

GitHub リポジトリを開く Creator のリポジトリを見る

ダウンロード

Manusで実行

Context Budget

More context is not better. Past a threshold, accuracy falls off a cliff (context rot).

Don't dump whole files — read the relevant function/section, not the 2000-line module.
Summarize, don't accumulate — replace a 200K-token transcript with a 4K recap of the load-bearing facts.
Drop dead context — once a sub-task is done, its detail leaves the window. Keep the decision, drop the trace.
State on disk, not in context — progress goes to a file the next turn re-reads, not into an ever-growing prompt. A bloated standing context (CLAUDE.md) taxes every single turn. Trim weekly. If the agent is confidently wrong, suspect the context before the model.

このリポジトリの他の Skills

同じリポジトリ

evaluator-calibration

Archive228/loopkit

Calibrate a reviewer persona with few-shot rubric examples so skepticism stays consistent and doesn't drift lenient over long runs.

2026-07-10561

feature-list-json

Archive228/loopkit

Enumerate every end-to-end feature as strict JSON entries with passes:false, editable-passes-only discipline, and priority order. The ledger fresh-context sessions read to know what's done, what's next, and what they're forbidden to touch.

2026-07-10561

harness-stripping

Archive228/loopkit

Systematically remove one harness component at a time and measure impact, killing scaffolding that no longer earns its complexity.

2026-07-10561

init-script-contract

Archive228/loopkit

Author idempotent init.sh under 120s plus sibling test.sh, stop.sh, reset.sh, serve.sh with fixed names the harness relies on.

2026-07-10561

planner-spec-expand

Archive228/loopkit

Expand a 1-4 sentence product brief into a full spec with a design language, acceptance surface, and an ordered feature list.

2026-07-10561

progress-reading-protocol

Archive228/loopkit

Run the fixed 6-step session-opening sequence — pwd, read progress, git log, count remaining features, init.sh, smoke-test last feature — before touching any new work. The orientation ritual that lets fresh-context sessions reconstruct project state in under a minute.

2026-07-10561

name	context-budget
description	Keep the agent's context lean so accuracy doesn't collapse. Use on long sessions, big files, or when the agent starts hallucinating.
when_to_use	long session, huge file dump, agent forgetting earlier steps, accuracy dropping

Context Budget

More context is not better. Past a threshold, accuracy falls off a cliff (context rot).

Don't dump whole files — read the relevant function/section, not the 2000-line module.
Summarize, don't accumulate — replace a 200K-token transcript with a 4K recap of the load-bearing facts.
Drop dead context — once a sub-task is done, its detail leaves the window. Keep the decision, drop the trace.
State on disk, not in context — progress goes to a file the next turn re-reads, not into an ever-growing prompt. A bloated standing context (CLAUDE.md) taxes every single turn. Trim weekly. If the agent is confidently wrong, suspect the context before the model.