itchernetski/claude-code-token-watch

Claude Code Token Watch is a small set of three Claude Code hook scripts that spot token-heavy anti-patterns during a session and inject a system message advising corrective behavior (they do not block actions). It targets Claude Code users concerned about running up token-based 5-hour limits (Pro/Max) — mainly developers using the Claude Code CLI with hooks enabled. Key features: - Three hooks: Read (same file read 3+ times), Bash (3+ recursive searches), WebFetch (3+ fetched pages). - Per-session counting using /tmp/claude-token-watch/<session_id>/; on the 3rd hit a systemMessage is emitted so the model usually self-corrects next turn. - Simple install: copy scripts to ~/.claude/hooks/, add entries to ~/.claude/settings.json, reload hooks. - Requires bash and jq; tested on macOS (Linux likely OK). Thresholds, messages and disabling are configurable. License: MIT.

AI-related keywords found in this repo's description, topics, or README summary — grouped by category. Each badge links to the corresponding ranking detail page.