agentleFS
Sign inStart free
binay-pybinay-py/claude-skills/skills/read-guard/SKILL.md

read-guard

Skill1 starsChanged 6 days agoMITView on GitHub

Stop Claude Code from filling its context window with whole-file reads of huge files and log dumps. A PreToolUse hook that denies Read of files over ~800 lines / 80 KB without offset/limit, and `cat` / `type` / `Get-Content` of big files with nothing limiting the output (cd-aware), telling Claude to Grep and read a slice instead. Retrying the exact same call is allowed, so nothing is ever blocked for good. Use when the user hits usage limits fast, when claude-token-audit shows Read or Bash output dominating, or when they ask to make Claude Code cheaper.

The file

Create a free account to mount
Files
SKILL.md

read-guard

scripts/read-guard.mjs is zero-dependency Node 22+. Every whole-file read of a 3,000-line file puts ~40k tokens into the context, and every later turn re-reads them. That's usually the single biggest avoidable cost (claude-token-audit shows it per tool).

Install (ask the user first, it changes their settings)

Merge into ~/.claude/settings.json (or a project's .claude/settings.json), using the absolute path:

{ "hooks": { "PreToolUse": [
  { "matcher": "Read|Bash|PowerShell",
    "hooks": [{ "type": "command", "command": "node \"/ABSOLUTE/PATH/read-guard/scripts/read-guard.mjs\" hook" }] }
] } }

What it does

CallDecision
Read of a text file > 800 lines or > 80 KB, without limitdeny. Claude is told the size and token cost and to Grep and then Read with offset/limit
Read with limit ≤ 800, images, PDFs, notebooks, binariesallow
cat / type / more / Get-Content of a big file (follows cd, Git Bash /c/... paths, 2>&1)deny, with suggestions to use grep -n, head, tail
the same command piped into another program (head, grep, jq, node script.js…, but not tee/cat/more), redirected to a file, or Get-Content -Tail/-TotalCountallow
the exact same call repeated within 10 minutesallow (escape hatch: a file that really must be read whole costs one extra turn, never a dead end)
anything else, or any error in the hookallow (fails open)

Thresholds: READ_GUARD_MAX_LINES (default 800) and READ_GUARD_MAX_KB (default 80).

node read-guard.mjs test path/to/file [--offset 100 --limit 200]
node read-guard.mjs test-cmd "cat server.log"

Limits

  • It can't shrink output from other commands (test runners, builds). Tell Claude to pipe noisy commands through tail -50 or use quiet flags.
  • Deny reasons go to Claude, not to the user. The user only sees a slightly different tool call.
  • Claude's Edit tool needs the file to have been read first. If a targeted Read isn't enough for an edit, Claude simply repeats the full Read and the escape hatch allows it.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

Check your setup, free