{
  "capabilities": [
    "tool:file_manager"
  ]
}

{
  "ok": true,
  "equipped": [
    {
      "capability": "tool:file_manager",
      "instruction": "Manage text files and directories inside the active Zuza runtime workspace.\n\nCall tool id `file_manager` through the provider tool interface with arguments\nmatching this schema.\n\nValid call shape is always:\n{\n  \"operations\": [\n    {\"action\": \"read_file\", \"path\": \"example.txt\"}\n  ]\n}\n\nThe tool id is `file_manager`. Never call `file_manager.write_file`,\n`file_manager.read_file`, or any other dotted method name.\n\nDo not send `operation`, top-level `action`, or top-level `path`; put each file\noperation inside `operations`.\nDo not send shell command strings such as `command`; use `list_directory`,\n`read_file`, or another supported action.\n\nUse this tool instead of `python_console` for ordinary file operations. For\nlarge files, use line-window `read_file` calls and exact-match `edit_file`\ncalls instead of writing Python scripts to slice, search, or replace text.\n\nFor generated reports or other artifacts, write the content directly when it is\nready. Use `append_file` when the task naturally produces later sections. Use\n`edit_file` for a targeted change to an existing file instead of reading and\nrewriting the whole file. Do not split solely to satisfy an arbitrary character\nlimit.\n\n`content` is mandatory for every `write_file` and `append_file` operation.\nCalls that omit it fail without modifying the target. Send an explicit empty\nstring only when intentionally creating or replacing a file with no text.\n\nSupported actions:\n- `create_directory`\n- `read_file`\n- `write_file`\n- `append_file`\n- `edit_file`\n- `list_directory`\n- `exists`\n\nInput contract:\n{\n  \"operations\": [\n    {\"action\": \"create_directory\", \"path\": \"./notes/research\"},\n    {\"action\": \"write_file\", \"path\": \"./notes/research/summary.md\", \"content\": \"# Summary\", \"overwrite\": true},\n    {\"action\": \"append_file\", \"path\": \"./notes/research/summary.md\", \"content\": \"\\n\\nMore notes.\"},\n    {\"action\": \"read_file\", \"path\": \"./notes/research/summary.md\", \"start_line\": 1, \"end_line\": 80},\n    {\"action\": \"edit_file\", \"path\": \"./notes/research/summary.md\", \"old_text\": \"# Summary\", \"new_text\": \"# Research summary\"},\n    {\"action\": \"list_directory\", \"path\": \"./notes\", \"recursive\": true, \"max_entries\": 50},\n    {\"action\": \"exists\", \"path\": \"./notes/research/summary.md\"}\n  ]\n}\n\nBehavior notes:\n- Paths are relative to the active runtime workspace. In a MiniZuza, this means\n  that MiniZuza's isolated workspace, not the Project Zuza application source.\n- A bare name, `.`, or a path beginning `./` resolves inside the\n  runtime-provided current working folder. Use `.` to list that folder.\n- Any other multi-part path starts at the workspace root. Examples include\n  `artifacts/2026-07-23/chat_12/report.md`, `uploads/source.pdf`, and\n  `tools/example/main.py`.\n- Never use a host absolute path, `file://` URL, browser URL, or display label\n  as a file address.\n- Reads and writes are UTF-8 text only.\n- The tool does not delete files or directories.\n- Every `read_file` call includes 1-based inclusive `start_line` and `end_line`.\n  To inspect a file from the beginning, start at line 1 and continue from\n  `next_start_line` only while the result's `has_more` is true. The result also\n  reports `total_lines`, so use the smallest focused range that can answer the\n  question or establish an exact edit.\n  Use `max_chars` only when a character boundary is genuinely useful; it may\n  stop inside a line.\n- `write_file` creates parent directories when needed.\n- `append_file` creates the file if it does not exist.\n- `edit_file` is the safe partial-write operation. It requires exact\n  `old_text` and `new_text` in the same operation object, changes nothing when\n  the old text is absent, and\n  changes nothing when the match is ambiguous. Include more surrounding text\n  to make the match unique. Set `replace_all: true` only when every exact match\n  should change.\n- Successful writes, appends, and edits report added and removed line counts.\n  Zuza stores the full unified diff outside model context for user review in\n  the chat interface.\n- General task deliverables default into the current working folder.\n- Every completed operation returns the canonical workspace-root-relative file\n  or directory address as `relative_path`. Copy that exact returned value for\n  later reads, edits, conversions, or `respond.attached_files`.\n- Use real source paths only when you are intentionally editing project files or\n  creating tools, senses, templates, tests, or project documentation.\n\nOutput contract:\n{\n  \"ok\": true,\n  \"tool\": \"file_manager\",\n  \"results\": []\n}"
    }
  ]
}

{
  "operations": [
    {
      "action": "read_file",
      "end_line": 6,
      "path": "admin-bounded-live-test.txt",
      "start_line": 4
    }
  ]
}

{
  "ok": true,
  "results": [
    {
      "ok": true,
      "action": "read_file",
      "relative_path": "artifacts/2026-08-17/chat_91/admin-bounded-live-test.txt",
      "content": "window-start\nMode: preview\nwindow-end",
      "truncated": false,
      "total_chars": 100,
      "selected_chars": 37,
      "total_lines": 10,
      "start_line": 4,
      "end_line": 6,
      "has_more": true,
      "next_start_line": 7
    }
  ]
}