Document Converter Workflow

Use this workflow when the user asks to read, summarize, analyze, or convert a large or complex document.

Do not start by converting the entire file into one large Markdown document. First build a lightweight understanding of the document, sample a few representative units, then decide how to read the rest.

What Conversion Means

This skill uses "conversion" in two document-specific ways:

• Semantic conversion: turn complex documents into model-readable Markdown artifacts, such as document.md, chunks/, document.index.json, document.outline.md, document.reading_state.json, visual-understanding records, and summaries. Use this for normal reading, analysis, summarization, and Markdown export.
• Format conversion: create a new file in another supported document format. Use this only when the user explicitly asks for a converted file or when a format mismatch must be repaired before extraction.

Current supported raw format conversions:

• PDF -> png, jpg, jpeg
• Office-like documents -> pdf, docx, pptx, xlsx

Office-like documents include Word, PowerPoint, spreadsheets, WPS Office, OpenDocument, RTF, templates, macro-enabled files, and slide-show files when the current runtime converter can handle them.

Other document types should usually be read or exported through semantic conversion, especially export_document_markdown, rather than forced through convert_document_format.

Default Approach

1. Inspect the document first: identify its type, size, structure, outline, and representative samples.
2. If the file is small, export it directly as one readable Markdown file.
3. If the file is large, sample a few representative units before choosing a full reading strategy.
4. Plan the next read from the sample result and the user's goal.
5. After inspection and sampling, use the returned strategy_context and the user's goal to choose an asset strategy.
6. If the user did not specify an asset preference and the recommendation materially changes the output, call ask_user once before extraction. Put the recommended strategy first and omit timeout; the user may need time to decide. Use timeout=60 only when the choice is simple, has a safe default, and does not materially affect the result.
7. Extract or understand only the selected ranges into reusable Markdown artifacts.
8. Read document.reading_state.json between large-document steps to avoid rereading the same content.
9. For summaries, summarize smaller chunks first, then combine them into section-level and document-level summaries.
10. Only perform format conversion when the user explicitly asks for a converted file or when conversion is necessary for extraction.

Large Document Rules

• Do not read or convert a large document all at once unless the user explicitly requires a full export.
• Always sample first for large files, then choose the next batch from the sample result.
• Use the outline, samples, and document.reading_state.json to decide what to read next.
• Prefer text extraction for ordinary document body content.
• Do not treat the presence of images as a reason to extract every embedded image object.
• Asset strategies are text_only, page_visual, reusable_assets, full_fidelity, and auto.
• Choose text_only for text reading, page_visual for slides/layout/charts, reusable_assets for standalone pictures or logos, and full_fidelity only when the user explicitly needs every source asset.
• When strategy_context.requires_user_confirmation is true and the user has not already specified the strategy, use one ask_user select question. The recommended strategy must be the first option and the default.
• Do not use one strategy blindly for the whole file. If the sample shows extractable text, read text ranges. If the sample is image-dominant or scanned, understand document images in batches.
• Do not call the generic visual understanding tool directly for document parsing. Use understand_document_images so results are saved and written back to chunks.
• When image understanding is used, process at most 10 images per call. Successful results are written back to the related chunk, and visual-results/ is preserved as the per-image recognition record.
• Before starting a large visual-understanding workload, call ask_user to ask whether the user wants to continue because it may take a long time. This applies when many batches are needed, such as scanned PDFs or slide decks with many image-only pages.
• Do not call ask_user for small visual reads, such as a document with only a few pages or a single batch of up to 10 images that is clearly needed for the user's request.
• For spreadsheets, inspect sheets, headers, sample rows, and table size before extracting data.
• For slide decks, treat each slide as a natural unit.
• For Word-like documents, follow the heading structure when available.
• For notebooks, preserve cell order and cell type.

Small Document Rules

• Small files can use export_document_markdown directly. The tool chooses the output artifact mode automatically.
• For small files, prefer the simple output structure: document.md, plus assets/ when needed. If visual understanding runs, keep visual-results/ as the recognition record.
• Do not create samples, chunks, indexes, or reading state just to read a small file.
• A PDF, Word file, PowerPoint file, or image set with 10 or fewer pages/slides/images is usually small.
• If a PDF, Word file, PowerPoint file, or image set has more than 10 pages/slides/images, use the large-document workflow instead of expecting a flat one-file result.
• Text, Markdown, and HTML files are small when they fit comfortably within one normal chunk.
• Small scanned files can be extracted and visually understood directly when needed; do not call ask_user unless many visual-understanding batches are required.
• Large-document exports automatically use progressive artifacts.

When The User Wants A Summary

For a large document summary, do not send all extracted text into the model at once.

Use this sequence:

1. Inspect the structure.
2. Sample representative pages, slides, sheets, or sections.
3. Plan the next read from the sample and the user's goal.
4. Extract relevant chunks, or understand image pages in batches when the document is scanned.
5. Summarize each chunk.
6. Merge chunk summaries into section summaries.
7. Merge section summaries into a final answer.

Code Mode Use

Use Code Mode as the execution path for this workflow. Keep the user-facing response focused on what was inspected, what was extracted, where the readable result is, and what can be done next. Do not expose internal class names, package paths, implementation details, or raw metadata unless the user asks for debugging details.

All document-converter SDK tool calls must be executed through the run_sdk_snippet tool by passing code in its python_code parameter.

Inside that python_code, use from sdk.tool import tool and call document-converter tools with tool.call(...). Always check result.ok before using the result.

Do not use run_python_snippet for document-converter tool calls. run_python_snippet is only a generic Python executor; it is not the intended execution path for Code Mode tools that require sdk.tool.call(...).

When a code example below starts with run_sdk_snippet(...), call that tool directly. Do not paste that wrapper into run_python_snippet.

Correct execution shape:

run_sdk_snippet(python_code="""
from sdk.tool import tool

result = tool.call("inspect_document", {
    "input_path": "/absolute/path/to/uploads/report.pdf",
})
if not result.ok:
    raise SystemExit(result.content)
print(result.content)
""")

All path parameters passed to these tools must be absolute paths.

For the same source document, reuse one stable output_dir across all document-converter tool calls. Keep document.index.json, document.outline.md, document.reading_state.json, samples/, chunks/, visual-results/, summaries, converted files, and combined Markdown exports under that same directory unless the user explicitly asks for a different location.

Asset path contract:

• document.index.json.assets[].path is canonical relative to output_dir, such as assets/page_001.png.
• Images inside chunks/*.md are relative to the chunk file, such as ../assets/page_001.png.
• Images inside root-level document.md are relative to document.md, such as assets/page_001.png.
• When opening an image for visual understanding, resolve the canonical asset path from output_dir; do not infer the filesystem path from the Markdown link text.

When the user does not provide an output directory, create one under the current workspace output area and derive the directory name from the current full file name, including the extension. For example, use sample_pdf for sample.pdf, sample_docx for sample.docx, and README_md for README.md. Do not use only the file stem in batch jobs, because files such as sample.pdf, sample.docx, and sample.png would otherwise overwrite or mix artifacts in the same sample/ directory.

Before writing to an existing output_dir, make sure it belongs to the same source file. If unsure, choose a fresh unique directory instead of reusing a possibly polluted directory.

Reuse Existing Parsed Artifacts

When starting a new topic or when the current context does not show whether a document has already been parsed, do not immediately re-run extraction. If the source file still exists, first look for an existing output directory that matches the current full file name, such as document-converter-output/sample_pdf/ for sample.pdf.

Reuse existing artifacts when all of these are true:

• document.index.json exists.
• document.outline.md exists.
• chunks/ exists and contains Markdown chunk files.
• The source_path in document.index.json points to the same source document.
• The source document has not changed since the index was created, when timestamps or stored metadata are available.

If valid parsed artifacts already exist, read document.reading_state.json when present, then read document.outline.md and document.index.json. Read only the needed chunk files or call summarize_document on that output_dir. Do not re-run export_document_markdown or extract_document_content just to rediscover content that is already available.

If only samples/ and document.reading_state.json exist, treat the document as sampled but not fully extracted. Use plan_document_reading to choose the next bounded extraction or image-understanding step.

Re-run extraction only when artifacts are missing, incomplete, point to a different source file, appear stale, do not cover the needed range, or the user explicitly asks to regenerate the Markdown.

When the user chooses a different explicit asset_strategy for a source that already has indexed artifacts, keep the original output_dir as the base name. The tool creates or reuses a sibling strategy version such as report_pdf__page_visual/; read the returned output_dir instead of assuming the base directory was overwritten.

Batch Conversion Rules

• Multi-file conversion is Code Mode orchestration, not a multi-file tool API. In batch work, run one run_sdk_snippet loop and call single-file tools such as inspect_document, export_document_markdown, and convert_document_format once per source file.
• Do not pass files, file_paths, multiple input_path values, or any invented batch parameter to document-converter tools. Their API boundary is one source document per tool call.
• Process batch inputs sequentially in list order. Do not use concurrent execution patterns in document-converter batch examples.
• Process every file requested by the user. Do not invent a local supported-extension whitelist.
• Office-like documents may include Microsoft Office, WPS Office, OpenDocument, RTF, templates, macro-enabled files, and slide-show files. Try the document-converter tools first and rely on the tool result to decide whether a specific file can be parsed in the current environment.
• Prefer calling inspect_document or export_document_markdown and record the returned error when a file is unsupported.
• Do not skip images, CSS, logs, configuration files, code files, or small text-like files just because they look different from office documents.
• Keep the source directory clean. Write reports, summaries, indexes, chunks, and combined Markdown files under the chosen output root, not inside the input directory.
• Use a unique output directory for each input file. The directory name must include the original extension as described above so that files such as sample.pdf, sample.docx, and sample.png do not mix artifacts in the same directory.
• For batch reports, record one row per file with input_path, output_dir, ok, file_type, index_path, combined_path, and error when present.
• Count success from actual tool results and output files, not from the planned file list.

Tools

inspect_document

Use this first for any large or complex document. It performs a lightweight inspection and returns the document type, scale, structure unit, outline, samples, and recommended strategy.

Parameters:

• input_path (required): absolute path to the source document.

Do not use this tool to read full content or create Markdown chunks.

sample_document_content

Use this after inspection for large or complex documents. It writes a small Markdown sample under samples/ and updates document.reading_state.json. Use the sample to decide whether the rest should be read as text, spreadsheet ranges, slides, or document images.

Parameters:

• input_path (required): absolute path to the source document.
• output_dir (required): absolute path to the stable output directory for this source document.
• ranges (optional): range expression such as 1-3,8.

Do not treat a sample as the final extraction.

plan_document_reading

Use this after sampling or partial extraction to decide the next bounded read.

Parameters:

• output_dir (required): absolute path to the stable output directory containing document-converter artifacts.
• goal (optional): current user goal, such as summary, clause lookup, or extracting decisions.

This tool only recommends the next action. It does not extract or summarize content.

build_document_index

Use this after inspection or sampling when you need a stable navigation map before extracting detailed content. It creates document.index.json and document.outline.md.

Parameters:

• input_path (required): absolute path to the source document.
• output_dir (required): absolute path to the directory where index and outline files should be written.

Do not use this tool for content extraction or summarization.

extract_document_content

Use this to read selected parts of a document into bounded Markdown chunks. Prefer targeted ranges over full-document extraction.

Parameters:

• input_path (required): absolute path to the source document.
• output_dir (required): absolute path to the directory where chunks/, document.index.json, and document.outline.md should be written.
• ranges (optional): range expression for the needed content, such as 1-3,8,10-12. The range may refer to pages, slides, sections, sheets, or cells depending on file type.
• asset_strategy (optional): auto, text_only, page_visual, reusable_assets, or full_fidelity. Use the sampled recommendation or the user's confirmed choice.

For PDF page_visual and full_fidelity, pass the whole requested range once. The tool automatically renders page snapshots in batches of at most 20 pages and combines the batches into one consistent export.

Do not use this as a blind whole-file converter unless the user explicitly requires a full export.

understand_document_images

Use this when extracted chunks show image-only pages, scans, charts, signatures, stamps, or other visual content needed for the user's goal. It understands document image assets and writes the result back into the related chunk.

This tool requires document.index.json and image assets under the same output_dir. If the document has only been sampled, first call extract_document_content for the selected range with image extraction enabled, then call this tool.

Parameters:

• output_dir (required): absolute path to the stable output directory containing document.index.json.
• ranges (optional): page or slide range whose images should be processed, such as 1-10. Omit it to process the next unread image batch.

Use this instead of the generic visual understanding tool when parsing document images.

export_document_markdown

Use this for the common request "convert this document to Markdown." For large files, inspect and sample first, then export the needed range or let the tool create progressive artifacts. It extracts the requested range, writes bounded chunk files when needed, updates the index and outline for progressive outputs, and writes document.md for user reading.

Parameters:

• input_path (required): absolute path to the source document.
• output_dir (required): absolute path to the directory where Markdown export artifacts should be written.
• ranges (optional): range expression to export. Omit it when the user explicitly wants the whole document.
• asset_strategy (optional): auto, text_only, page_visual, reusable_assets, or full_fidelity. Use the sampled recommendation or the user's confirmed choice.

For PDF page_visual and full_fidelity, page snapshots are rendered in batches of at most 20 pages. The result metadata includes execution_batches; do not wait for a limit error and retry manually.

This tool is a convenience workflow for Markdown export, not a replacement for targeted extraction when the user only needs specific content.

For small files, the main artifact is document.md. Do not expect chunks/, document.index.json, document.outline.md, or document.reading_state.json to exist.

In batch exports, output_dir must be unique for each source file and should include the source file extension in the directory name.

summarize_document

Use this after chunks already exist. It summarizes from document.index.json and chunks/ instead of reading the original document again.

Parameters:

• output_dir (required): absolute path to the directory containing document.index.json and chunks/.

Do not call this before extracting content.

convert_document_format

Use this only when the user explicitly asks for a converted file or conversion is required before extraction. It only changes file format.

This tool is not a universal file converter. It supports the raw format conversion matrix described in "What Conversion Means". If a requested route is unsupported, use the tool error to decide whether to switch to export_document_markdown, correct the requested target format, or tell the user that this document-to-document conversion is not currently supported.

After this tool succeeds, continue with the converted file path returned in result.data["output_files"][0]. Do not continue parsing, inspecting, extracting, or exporting with the original input_path, because the original file may still have the same format mismatch that required conversion.

Parameters:

• input_path (required): absolute path to the source document.
• output_dir (required): absolute path to the directory for converted files.
• target_format (required): target format, such as pdf, png, jpg, docx, pptx, or xlsx.
• ranges (optional): page range for conversion tasks that support ranges, such as rendering selected PDF pages to images.

Do not use this tool for semantic extraction, chunking, indexing, or summarization.

If this tool is used to fix a format mismatch, the next document-converter call should use the converted path:

• inspect_document.input_path = converted.data["output_files"][0]
• export_document_markdown.input_path = converted.data["output_files"][0]
• extract_document_content.input_path = converted.data["output_files"][0]

Code Mode Examples

Sample and plan before reading a large document

run_sdk_snippet(python_code="""
import re
from pathlib import Path
from sdk.tool import tool

def document_output_dir(output_root: str, input_path: str) -> str:
    path = Path(input_path)
    safe_name = re.sub(r"[^\w.-]+", "_", path.name, flags=re.UNICODE).strip("._")
    safe_name = safe_name.replace(".", "_") or "document"
    return str(Path(output_root) / safe_name)

doc = "/absolute/path/to/uploads/report.pdf"
out = document_output_dir("/absolute/path/to/document-output", doc)

profile = tool.call("inspect_document", {
    "input_path": doc,
})
if not profile.ok:
    raise SystemExit(profile.content)

sample = tool.call("sample_document_content", {
    "input_path": doc,
    "output_dir": out,
})
if not sample.ok:
    raise SystemExit(sample.content)

plan = tool.call("plan_document_reading", {
    "output_dir": out,
    "goal": "summarize the document",
})
if not plan.ok:
    raise SystemExit(plan.content)

print(plan.content)
""")

Extract only the needed range

run_sdk_snippet(python_code="""
from sdk.tool import tool

doc = "/absolute/path/to/uploads/report.pdf"
out = "/absolute/path/to/document-output/report_pdf"

result = tool.call("extract_document_content", {
    "input_path": doc,
    "output_dir": out,
    "ranges": "1-3,8,10-12",
})
if not result.ok:
    raise SystemExit(result.content)

print(result.content)
""")

Use understand_document_images for image-only pages after chunks and image assets exist.

Understand scanned or image-only pages

run_sdk_snippet(python_code="""
from sdk.tool import tool

doc = "/absolute/path/to/uploads/report.pdf"
out = "/absolute/path/to/document-output/report_pdf"

extract = tool.call("extract_document_content", {
    "input_path": doc,
    "output_dir": out,
    "ranges": "1-10",
})
if not extract.ok:
    raise SystemExit(extract.content)

result = tool.call("understand_document_images", {
    "output_dir": out,
    "ranges": "1-10",
})
if not result.ok:
    raise SystemExit(result.content)

print(result.content)
""")

Export a document to Markdown

run_sdk_snippet(python_code="""
import re
from pathlib import Path
from sdk.tool import tool

def document_output_dir(output_root: str, input_path: str) -> str:
    path = Path(input_path)
    safe_name = re.sub(r"[^\w.-]+", "_", path.name, flags=re.UNICODE).strip("._")
    safe_name = safe_name.replace(".", "_") or "document"
    return str(Path(output_root) / safe_name)

doc = "/absolute/path/to/uploads/report.pdf"
out = document_output_dir("/absolute/path/to/document-output", doc)

export = tool.call("export_document_markdown", {
    "input_path": doc,
    "output_dir": out,
})
if not export.ok:
    raise SystemExit(export.content)

print(export.content)
""")

Export multiple documents to Markdown sequentially

run_sdk_snippet(python_code="""
import json
import re
from pathlib import Path
from sdk.tool import tool

def document_output_dir(output_root: str, input_path: str) -> str:
    path = Path(input_path)
    safe_name = re.sub(r"[^\\w.-]+", "_", path.name, flags=re.UNICODE).strip("._")
    safe_name = safe_name.replace(".", "_") or "document"
    return str(Path(output_root) / safe_name)

docs = [
    "/absolute/path/to/uploads/report.pdf",
    "/absolute/path/to/uploads/minutes.docx",
    "/absolute/path/to/uploads/slides.pptx",
]
output_root = "/absolute/path/to/document-output"

reports = []
for doc in docs:
    out = document_output_dir(output_root, doc)
    export = tool.call("export_document_markdown", {
        "input_path": doc,
        "output_dir": out,
    })
    reports.append({
        "input_path": doc,
        "output_dir": out,
        "ok": export.ok,
        "file_type": export.data.get("file_type"),
        "index_path": export.data.get("index_path"),
        "combined_path": export.data.get("combined_path"),
        "error": None if export.ok else export.content,
    })

print(json.dumps(reports, ensure_ascii=False, indent=2))
""")

Summarize from the index and chunks

run_sdk_snippet(python_code="""
from sdk.tool import tool

summary = tool.call("summarize_document", {
    "output_dir": "/absolute/path/to/document-output/report",
})
if not summary.ok:
    raise SystemExit(summary.content)

print(summary.content)
""")

Convert format only when conversion is the user's goal

run_sdk_snippet(python_code="""
from sdk.tool import tool

converted = tool.call("convert_document_format", {
    "input_path": "/absolute/path/to/uploads/slides.pptx",
    "output_dir": "/absolute/path/to/document-output/slides",
    "target_format": "pdf",
})
if not converted.ok:
    raise SystemExit(converted.content)

print(converted.content)
""")

Convert first, then export with the converted path

Use this pattern when inspect_document reports a file format mismatch, such as a file named .docx whose real content is .docm.

run_sdk_snippet(python_code="""
from sdk.tool import tool

original_doc = "/absolute/path/to/uploads/report.docx"
out = "/absolute/path/to/document-output/report_docx"

converted = tool.call("convert_document_format", {
    "input_path": original_doc,
    "output_dir": out,
    "target_format": "docx",
})
if not converted.ok:
    raise SystemExit(converted.content)

converted_doc = converted.data["output_files"][0]

export = tool.call("export_document_markdown", {
    "input_path": converted_doc,
    "output_dir": out,
})
if not export.ok:
    raise SystemExit(export.content)

print(export.content)
""")