Built a local pre-processing plugin for Claude Code after research & measuring real metrics (3,500+ tests, v1.22.1)

Описание и применение

Создан локальный плагин chamnan для Claude Code, обеспечивающий детерминированное обрезание контекста, локальную очистку секретов и сохранение истории сессий, что повышает предсказуемость и безопасность работы с большими репозиториями.

Задача

Large, messy repositories cause unpredictable context usage, token consumption, and potential data leaks when using Claude Code.

Результат

The plugin dramatically reduces Quick Index overhead, prevents accidental data leaks, and preserves session context across runs, enabling more reliable and secure Claude Code usage on large, complex repositories.

Рабочий процесс

• 1. Scan repository and pre-index files to determine context size.
• 2. Trim or split files to fit within the token budget before sending to Claude Code.
• 3. Perform local redaction of secrets and internal strings.
• 4. Store session decisions and project history in .chamnan/ for continuity.
• 5. Pass the sanitized, trimmed context to Claude Code for processing.

Доказательства

Built in pure Python (standard library only, zero external packages) and backed by 3,500+ automated tests.

Источник

Deterministic Context Budgeting: Pre-indexes and trims repository context before passing it to Claude Code (slashed Quick Index overhead dramatically in real-world tests).

Источник