Summarise Large Files with Gemini 2.5 Flash

#!/usr/bin/env bash
#
# summarise-file - Summarise large files using Gemini 2.5 Flash
#
# Usage:
#   summarise-file <file>           # Summarise a file
#   summarise-file -                # Read from stdin
#   cat file.txt | summarise-file   # Pipe content
#
# Requires:
#   GEMINI_API_KEY environment variable

set -euo pipefail

MODEL="gemini-2.5-flash"
API_URL="https://generativelanguage.googleapis.com/v1beta/models/${MODEL}:generateContent"

# Check for API key
if [[ -z "${GEMINI_API_KEY:-}" ]]; then
    echo "Error: GEMINI_API_KEY not set" >&2
    echo "Get one at https://aistudio.google.com/apikey" >&2
    exit 1
fi

# Get content from file or stdin
if [[ $# -eq 0 ]] || [[ "$1" == "-" ]]; then
    CONTENT=$(cat)
    FILENAME="stdin"
elif [[ -f "$1" ]]; then
    CONTENT=$(cat "$1")
    FILENAME=$(basename "$1")
else
    echo "Error: File not found: $1" >&2
    exit 1
fi

# Check content isn't empty
if [[ -z "$CONTENT" ]]; then
    echo "Error: No content to summarise" >&2
    exit 1
fi

# Get file size for context
CHAR_COUNT=${#CONTENT}
if [[ $CHAR_COUNT -gt 1000000 ]]; then
    SIZE_NOTE="(~$(( CHAR_COUNT / 1000 ))k chars)"
else
    SIZE_NOTE="(${CHAR_COUNT} chars)"
fi

echo "Summarising ${FILENAME} ${SIZE_NOTE}..." >&2

# Build JSON payload - pipe content to jq to avoid argument limits
SYSTEM_PROMPT="Summarise the following document concisely. Focus on key points, main arguments, and actionable information. If it's code, describe what it does. If it's structured data, describe the schema and notable entries.

---

"

PAYLOAD=$(printf '%s%s' "$SYSTEM_PROMPT" "$CONTENT" | jq -Rs '{
    contents: [{parts: [{text: .}]}],
    generationConfig: {temperature: 0.3, maxOutputTokens: 2048}
}')

# Make API call (pipe payload to avoid argument length limits)
RESPONSE=$(echo "$PAYLOAD" | curl -s -X POST "${API_URL}?key=${GEMINI_API_KEY}" \
    -H "Content-Type: application/json" \
    --max-time 120 \
    --data-binary @-)

# Check for errors
if echo "$RESPONSE" | jq -e '.error' >/dev/null 2>&1; then
    ERROR_MSG=$(echo "$RESPONSE" | jq -r '.error.message')
    echo "API Error: ${ERROR_MSG}" >&2
    exit 1
fi

# Extract and print the summary
SUMMARY=$(echo "$RESPONSE" | jq -r '.candidates[0].content.parts[0].text // "No response generated"')

echo ""
echo "$SUMMARY"

chmod +x ~/.local/bin/summarise-file

# Add to BOTH if you use zsh but want Claude Code to use it
# (Claude Code runs in bash, not zsh)
echo 'export GEMINI_API_KEY="your-key-here"' >> ~/.zshrc
echo 'export GEMINI_API_KEY="your-key-here"' >> ~/.bashrc
source ~/.zshrc

summarise-file /path/to/large-file.txt

pdftotext document.pdf - | summarise-file

cat /var/log/syslog | summarise-file

#!/bin/bash
# PreToolUse hook for Read: block large files and suggest summarise-file
# Exit codes: 0=allow, 1=deny (ask user), 2=block with message

INPUT=$(cat)
FILE=$(echo "$INPUT" | jq -r '.tool_input.file_path // empty')

# Skip if no file path or file doesn't exist
[ -z "$FILE" ] && exit 0
[ ! -f "$FILE" ] && exit 0

# Get file size in bytes
SIZE=$(stat -c%s "$FILE" 2>/dev/null || echo 0)

# Threshold: 200KB (below Claude Code's 256KB internal limit)
THRESHOLD=204800

if [ "$SIZE" -gt "$THRESHOLD" ]; then
  SIZE_HUMAN=$(numfmt --to=iec "$SIZE")
  echo "File is ${SIZE_HUMAN}. Use instead: summarise-file \"$FILE\"" >&2
  exit 2  # BLOCK
fi

exit 0  # ALLOW

chmod +x ~/.claude/hooks/check-file-size.sh

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Read",
        "hooks": [
          {
            "type": "command",
            "command": "/home/YOUR_USERNAME/.claude/hooks/check-file-size.sh"
          }
        ]
      }
    ]
  }
}