mufarosean/scoped-extraction

Scoped Extraction Workflow

You are executing the scoped-extraction skill. Follow these steps precisely. Do not skip files — every file in scope must be checked.

Step 1: Parse the extraction request

From $ARGUMENTS, determine:

• What to extract (e.g. "deductible amounts", "contact names", "dates")
• Where to look (specific folder, file type, or entire workspace)

If the scope is unclear, default to the entire workspace.

Step 2: Enumerate files in scope

Use list_files to enumerate all files in the target scope. Record the complete file list as your coverage checklist.

Step 3: Read and extract from each file

For every file in the coverage checklist:

1. Use read_file to read the file's content.
2. Search for instances of the target information.
3. If found, record:
   • The extracted value(s)
   • The file path where it was found
   • The context (surrounding text, section heading)
4. If NOT found in this file, note: "No matching information in [file]"

Do NOT skip files. Even if you think a file is unlikely to contain the target information, read it and check.

Step 4: Aggregate results

Present the extracted information in a structured format:

1. Extraction target: What was searched for
2. Scope: What files/folders were searched
3. Results: Table or list with:
   • Value found
   • Source file (with path)
   • Context snippet
4. Files with no matches: List files that were checked but contained no relevant information
5. Coverage: "Checked X/Y files" (X must equal Y)

Step 5: Identify conflicts

If the same type of information has different values in different files:

• Flag the conflict prominently
• Show both values with their source files
• Do not silently pick one value over another

Important notes

• Completeness over speed — check every file, no exceptions
• Cite sources using [N] notation
• If a value appears approximate or uncertain, note the uncertainty
• Distinguish between "not found" (checked, absent) and "not checked" (should never happen)