File Grep

🔎 File Grep Skill

Overview

This skill enables the LLM to help users search for a text pattern inside files (basic grep-like functionality) by producing a strictly-formatted JSON output that describes the search. The JSON output is then parsed by the file_grep_execute skill to walk the filesystem, match the pattern, and return the list of matching file locations (and optionally the matching lines).

The default working directory is the current directory (./) unless the user specifies otherwise.

Skill Name

file-grep

When to Use This Skill

Invoke this skill whenever the user's request implies searching text inside files. Look for phrases such as:

• "grep for ..."
• "search for ... in files"
• "find files containing ..."
• "which files have ..."
• "look up ... in the codebase"
• "where is the string ... used"
• "list files with the word ..."

Do NOT use this skill when the user wants to:

• Create a file → use file-creator.
• Modify a file → use file-changer.
• Read the entire content of one specific file → use file-reader.

What the LLM Must Do

When the skill is triggered, the LLM must:

1. Extract the following from the user's request (asking clarifying questions only if truly ambiguous):

   • pattern (required): The string or regex to search for.
   • path (optional): The directory or file to search in. Defaults to ./ (current working directory).
   • recursive (optional): true to search subdirectories. Defaults to true.
   • isRegex (optional): true to interpret pattern as a Python regular expression. Defaults to false (plain substring search).
   • caseSensitive (optional): true for case-sensitive matching. Defaults to false.
   • includeGlob (optional): A glob pattern to restrict which files are searched (e.g. *.py, *.md). Defaults to * (all files).
   • excludeGlob (optional): Glob pattern(s) to skip (comma-separated). Defaults to common noise directories like .git,node_modules,__pycache__,.venv,dist,build.
   • maxResults (optional): Maximum number of matching files to return. Defaults to 100.
   • showLines (optional): true to include matching lines with line numbers in the output. Defaults to true.
   • sequential (optional): presented if the grep result is needed as input for a follow-up AI task. Mirrors the file-reader sequential object.
   • prompt (optional): When sequential is presented, this prompt is used to ask the AI to perform all of follow-up tasks on the grep result.
   • description (optional): One-line summary of the search.

2. Produce a response containing a single pure JSON object that STRICTLY follows the schema below. No code fences, no extra prose.

Required JSON Output Schema

The LLM's response MUST be a pure JSON string with no code blocks (no leading ```).

{
  "action": "grep_file",
  "thinking": "<ai-thinking-trace>",
  "sequential": {
    "prompt": ""
  },
  "properties": [
    { "name": "pattern",       "value": "<text-or-regex>" },
    { "name": "path",          "value": "./" },
    { "name": "recursive",     "value": "true" },
    { "name": "isRegex",       "value": "false" },
    { "name": "caseSensitive", "value": "false" },
    { "name": "includeGlob",   "value": "*" },
    { "name": "excludeGlob",   "value": ".git,node_modules,__pycache__,.venv,dist,build" },
    { "name": "maxResults",    "value": "100" },
    { "name": "showLines",     "value": "true" },
  ]
}

Field Rules

Formatting Requirements

• The JSON output must be valid JSON (parseable by json.loads).
• All string values must properly escape special characters (", \, newline as \n, tab as \t).
• Do NOT emit multiple file-grep payloads in a single response.
• If the user does not specify a pattern to search, do NOT execute this skill — ask for the pattern first or produce an empty action.

Response Pattern for the LLM

A typical assistant reply should look like the following. No other output is needed. Output must be in pure JSON format.

{ "action": "grep_file", "thinking": "User wants to find files that contain the word TODO in the current directory.", "properties": [ { "name": "pattern", "value": "TODO" }, { "name": "path", "value": "./" }, { "name": "recursive", "value": "true" }, { "name": "isRegex", "value": "false" }, { "name": "caseSensitive", "value": "false" }, { "name": "includeGlob", "value": "*" }, { "name": "excludeGlob", "value": ".git,node_modules,pycache,.venv,dist,build" }, { "name": "maxResults", "value": "100" }, { "name": "showLines", "value": "true" }, { "name": "description", "value": "Search TODO markers in the project" } ] }

Decision Logic

User Prompt
    |
    +-- Does the request imply searching text inside files?
    |     |
    |     +-- YES --> Extract pattern/path/flags --> Emit ONE `file-grep` JSON payload
    |     |
    |     +-- NO  --> Do not use this skill

Examples

Example 1: Simple substring search in current directory

User: "Find all files containing def main here."

LLM emits: { "action": "grep_file", "thinking": "User wants a plain substring search for 'def main' in the current directory, recursively.", "properties": [ { "name": "pattern", "value": "def main" }, { "name": "path", "value": "./" }, { "name": "recursive", "value": "true" }, { "name": "isRegex", "value": "false" }, { "name": "caseSensitive", "value": "false" }, { "name": "includeGlob", "value": "*" }, { "name": "excludeGlob", "value": ".git,node_modules,pycache,.venv,dist,build" }, { "name": "maxResults", "value": "100" }, { "name": "showLines", "value": "true" }, { "name": "description", "value": "Locate main function definitions" } ] }

Example 2: Case-sensitive search restricted to Python files

User: "Which .py files contain the class MyClass? Be case sensitive."

LLM emits: { "action": "grep_file", "thinking": "User asked case-sensitive search restricted to .py files.", "properties": [ { "name": "pattern", "value": "MyClass" }, { "name": "path", "value": "./" }, { "name": "recursive", "value": "true" }, { "name": "isRegex", "value": "false" }, { "name": "caseSensitive", "value": "true" }, { "name": "includeGlob", "value": "*.py" }, { "name": "excludeGlob", "value": ".git,node_modules,pycache,.venv,dist,build" }, { "name": "maxResults", "value": "100" }, { "name": "showLines", "value": "true" }, { "name": "description", "value": "Find Python files defining MyClass" } ] }

Example 3: Regex search in a specific folder

User: "In src/, find files matching the regex TODO\\(.*\\)."

LLM emits: { "action": "grep_file", "thinking": "Regex search restricted to src/ folder.", "properties": [ { "name": "pattern", "value": "TODO\(.\)" }, { "name": "path", "value": "src/" }, { "name": "recursive", "value": "true" }, { "name": "isRegex", "value": "true" }, { "name": "caseSensitive", "value": "false" }, { "name": "includeGlob", "value": "" }, { "name": "excludeGlob", "value": ".git,node_modules,pycache,.venv,dist,build" }, { "name": "maxResults", "value": "100" }, { "name": "showLines", "value": "true" }, { "name": "description", "value": "Regex TODO scan in src/" } ] }

Example 4: File-name-only output

User: "Just list the files that contain the word password, don't show lines."

LLM emits:(no code fence)

{
  "action": "grep_file",
  "thinking": "User only wants file paths, not the matched lines.",
  "properties": [
    { "name": "pattern", "value": "password" },
    { "name": "path", "value": "./" },
    { "name": "recursive", "value": "true" },
    { "name": "isRegex", "value": "false" },
    { "name": "caseSensitive", "value": "false" },
    { "name": "includeGlob", "value": "*" },
    { "name": "excludeGlob", "value": ".git,node_modules,__pycache__,.venv,dist,build" },
    { "name": "maxResults", "value": "100" },
    { "name": "showLines", "value": "false" },
    { "name": "description", "value": "List files containing 'password'" }
  ]
}

Safety Guidelines

1. Never search paths outside the user's intent (no arbitrary system paths).
2. Refuse suspicious search roots (e.g., C:\Windows\, /etc/, ~/.ssh/).
3. Cap the number of files returned via maxResults to avoid runaway output.
4. Skip binary files and unreadable files silently — do not crash.
5. Always echo the JSON payload so the user can review before execution.