files.read

Schema

Input = {
  path: string,
  maxChars?: number,      // 100-2000000, default 200000
  base64?: boolean,        // default false — force base64 even for text
  startLine?: number,       // 1-based
  endLine?: number,
}

Output = {
  path: string,
  mime: string,
  encoding: 'utf8' | 'base64',
  content: string,
  size: number,
  truncated: boolean,
  lines?: number,           // total line count, for text files
}

Binary detection is automatic: if base64 isn’t requested, the skill still switches to base64 encoding when the file’s extension doesn’t look textual and a byte-level sniff of the buffer thinks it’s binary. When startLine/endLine are given, only that slice of the file is returned (and still counted against maxChars) — useful for an agent reading one section of a long generated file without pulling the whole thing into context.

Example

curl -s -X POST http://localhost:4000/api/studio/skills/invoke \
  -H 'content-type: application/json' -b cookies.txt \
  -d '{"name":"files.read","input":{"path":"research/briefing.md","maxChars":2000}}'

{
  "ok": true,
  "output": {
    "path": "research/briefing.md",
    "mime": "text/markdown",
    "encoding": "utf8",
    "content": "# Research briefing\n\n...",
    "size": 4821,
    "truncated": false,
    "lines": 96
  }
}

Used everywhere a task needs a dependency’s earlier output as a real file, not just the truncated inline preview the runner already provides in-context.