| name | stingy-guard |
| description | Toggle the token efficiency guard hook on or off. When active, it monitors every
tool call and warns about token-wasteful patterns: reading entire large files,
using Bash instead of dedicated tools, unnecessary agent spawns, and more.
Use when: "stingy guard", "guard on", "guard off", "watch my tokens",
"monitor efficiency", "enable token guard", "disable token guard".
|
| allowed-tools | ["Bash","Read","AskUserQuestion"] |
/stingy-guard — Token Efficiency Guard
Toggle the pre-tool-use hook that monitors every tool call for token waste.
What It Does
When active, the guard intercepts tool calls BEFORE they execute and warns about:
- Read of files >500 lines without offset/limit
- Bash running
cat, grep, find when dedicated tools are cheaper
- Agent spawns (each duplicates ~15-30K tokens of system context)
- WebSearch when local knowledge might suffice
The guard only WARNS — it never blocks. You see the warning and decide whether
to proceed.
When to Recommend /clear
Beyond per-tool-call warnings, proactively suggest /clear when you notice:
- Context bloat after a completed task — if a multi-step task just finished (deploy, review, refactor) and the conversation is full of stale tool results, recommend
/clear before starting the next unrelated task. Old context wastes tokens on every subsequent message.
- Long agent result dumps — after receiving large agent results or multi-file diffs that won't be referenced again, suggest clearing to reclaim context.
- Topic switch — when the user starts a new, unrelated task in the same session. Prior context adds zero value but costs tokens on every turn.
- Repeated re-reads — if you notice the same files being read multiple times because earlier reads were compressed out of context, that's a sign the conversation is too long.
/clear and re-read once is cheaper than re-reading on every turn.
- Post-commit/deploy — after committing, pushing, deploying, and verifying, the conversation is full of diffs, build output, and test results that won't be needed again.
How to recommend: Don't be annoying about it. Mention it once at the end of a natural milestone:
💡 Good time to /clear — this task is done and the context is heavy. Fresh start will save tokens on your next ask.
Step 1: Check Current Status
~/.claude/skills/stingy-guard/bin/toggle-hook.sh status
Step 2: Toggle
Ask the user:
Token guard is currently [ACTIVE/INACTIVE].
A) Turn ON — warns before wasteful tool calls
B) Turn OFF — no monitoring
C) Show me what it checks
If A (enable):
~/.claude/skills/stingy-guard/bin/toggle-hook.sh on
If B (disable):
~/.claude/skills/stingy-guard/bin/toggle-hook.sh off
If C (show): explain each check pattern:
- Read >500 lines: Files over 500 lines without offset/limit waste tokens. Use targeted reads.
- Bash cat/grep/find: Dedicated Read/Grep/Glob tools are cheaper than shell wrappers.
- Agent spawns: Each agent duplicates ~15-30K tokens of system prompt + CLAUDE.md.
- WebSearch: Search results consume significant tokens. Check local files first.
Tell the user to restart Claude Code after toggling for the change to take effect.
Rules
- The toggle script backs up settings.json automatically before any change
- If JSON validation fails after editing, the backup is restored automatically
- The guard uses jq (preferred) or falls back to bash heuristics — no python3 required
- Warnings go to stderr so they don't interfere with tool output
