Line-Context File Reading

When language models inspect source code in tools like Claude Code or Codex, plain text is not enough. The model needs precise line references to reason about code blocks, report diagnostic locations, and calculate exact substitutions. For this reason, agent reading tools format content with 1-based line numbers and support slicing via start_line and end_line.

Input and Output Protocol

Your program reads a JSON object from standard input containing the operation and the in-memory workspace:

{
  "op": "read_file",
  "workspace": {
    "hello.py": "print('hello')\nprint('world')\n"
  },
  "path": "hello.py",
  "start_line": 1,
  "end_line": 2
}

When successful, write compact JSON to standard output with keys in alphabetical order:

{
  "content": "1: print('hello')\n2: print('world')\n",
  "status": "ok",
  "total_lines": 2
}

The keys are:

• content: string containing the requested lines, each prefixed with "{num}: {line}\n". If the file is empty or range has no lines, returns "".
• status: always "ok" on success.
• total_lines: integer representing the total line count of the file.

Error Handling

If the path does not exist in workspace, if op is not "read_file", or if standard input is malformed:

• Leave standard output empty.
• Write an error description to standard error containing the word error.
• Exit with status code 1.

Acceptance Criteria

• Formats all lines with 1-based prefixes when boundaries are omitted.
• Correctly filters lines when start_line and end_line are provided.
• Gracefully clamps boundaries when end_line exceeds total line count.
• Rejects missing files and invalid JSON with exit code 1.