{"id":144224,"date":"2026-05-09T05:32:47","date_gmt":"2026-05-09T05:32:47","guid":{"rendered":"\/ca\/tutorials\/extract-dates-from-pdfs-openclaw"},"modified":"2026-05-09T05:32:47","modified_gmt":"2026-05-09T05:32:47","slug":"extract-dates-from-pdfs-openclaw","status":"publish","type":"post","link":"\/ca\/tutorials\/extract-dates-from-pdfs-openclaw","title":{"rendered":"How to extract dates from PDFs using OpenClaw"},"content":{"rendered":"

To extract dates from PDFs using OpenClaw<\/strong>, upload or reference the PDF, run the PDF tool with the date extraction prompt, return the dates as structured JSON, and validate the results before exporting. The workflow works for invoices, contracts, reports, and other business documents that contain dates like invoice_date, due_date, effective_date, expiration_date, or issue_date.<\/p>

The basic workflow has five steps:<\/p>

    \n
  1. Configure OpenClaw with a PDF-capable model or use Managed OpenClaw for a 1-click setup.<\/li>\n\n\n\n
  2. Add the PDF to your workspace or provide a URL to an accessible file.<\/li>\n\n\n\n
  3. Ask OpenClaw to extract dates as structured JSON, with fields such as date_iso, date_raw, date_type, confidence, and source_page.<\/li>\n\n\n\n
  4. Use OCR or page targeting when the PDF is scanned, long, or only has dates on specific pages.<\/li>\n\n\n\n
  5. Validate ambiguous or low-confidence dates before exporting them.<\/li>\n<\/ol>

    <\/p>

    What do you need before extracting PDF dates with OpenClaw?<\/h2>

    Before extracting dates from PDFs with OpenClaw<\/a>, you need a working environment, access to a PDF-capable AI model, and a clear extraction schema for the dates you want to capture. These three parts let the OpenClaw PDF tool read the document, identify date values, and return them in a format your spreadsheet, CRM, accounting system, or database can use.<\/p>

    The fastest setup path is a managed OpenClaw<\/a> solution, which deploys OpenClaw in 1 click and includes built-in AI access. This option works well if you want to process invoices, contracts, or reports without manually configuring providers, maintaining a server, or keeping a local machine online.<\/p>

    A self-managed OpenClaw setup gives you more control. Choose this path if you need root access, custom OCR tools, modified OpenClaw skills, specific Python packages, or deeper integrations with internal systems.<\/p>

    Before running your first PDF date extraction, prepare the following:<\/p>

      \n
    1. A PDF file that OpenClaw can access. <\/strong>Place the PDF in your OpenClaw workspace or use an accessible file URL. For example, invoice PDFs can be stored in workspace\/invoices\/, while contract PDFs can be stored in workspace\/contracts\/.<\/li>\n\n\n\n
    2. A PDF-capable model. <\/strong>OpenClaw needs a model that can process PDF content through the PDF tool. Native PDF mode works with providers that support direct PDF reading, while fallback mode extracts the PDF content first and sends the extracted text or images to the selected model.<\/li>\n\n\n\n
    3. A configured PDF model setting. <\/strong>In a self-managed setup, set the default PDF model in your OpenClaw configuration, usually through a setting such as agents.defaults.pdfModel. In Managed OpenClaw, much of this setup is handled through the hosted environment.<\/li>\n\n\n\n
    4. PDF size and page limits. <\/strong>Check the maximum file size, maximum page count, and number of PDFs allowed per tool call before processing large batches. These limits matter because long reports, scanned contracts, and multi-file invoice batches may need to be split into smaller jobs.<\/li>\n\n\n\n
    5. A date extraction schema. <\/strong>Decide which date fields you want before writing the prompt. Common fields include invoice_date, due_date, effective_date, expiration_date, issue_date, and other. Also, decide whether the output should include date_iso, date_raw, confidence, hil_flag, and source_page.<\/li>\n\n\n\n
    6. A destination for the extracted dates. <\/strong>Choose where the structured output should go after extraction. For a simple workflow, save the result as JSON or CSV. For a business workflow, send it to Google Sheets, Airtable, a CRM, an accounting system, or a contract management database.<\/li>\n<\/ol>

      The most important requirement is not just that OpenClaw can read the PDF. The workflow also needs to return dates in a predictable structure. For example, asking OpenClaw to “find the dates” returns an answer that still needs manual cleanup. Asking OpenClaw for date_iso, date_raw, date_type, confidence, and source_page returns data that can be fed directly into an automated validation or export step.<\/p>

      After the setup is complete, the next step is writing the extraction prompt that tells OpenClaw which dates to extract and how to format the results.<\/p>

      How do you write a date extraction prompt for OpenClaw?<\/h2>

      A date extraction prompt specifies which dates to extract from a PDF, how to classify each date, and the output format to return. For automated workflows, the prompt should request structured JSON rather than a written summary, as JSON can be validated, filtered, and exported to a spreadsheet or database.<\/p>

      A strong OpenClaw date extraction prompt should include five instructions:<\/p>

        \n
      1. Define the task clearly. <\/strong>Tell OpenClaw to extract dates from the attached PDF, not to summarize the document or extract every field.<\/li>\n\n\n\n
      2. Specify the date types. <\/strong>List the business date categories you need, such as invoice_date, due_date, effective_date, expiration_date, issue_date, and others.<\/li>\n\n\n\n
      3. Request normalized and raw values. <\/strong>Ask for both date_iso and date_raw. The ISO value provides a clean date for storage, while the raw value supports auditing and human review.<\/li>\n\n\n\n
      4. Add confidence and review fields.<\/strong> Include confidence and hil_flag so the workflow can separate reliable dates from values that need a human check.<\/li>\n\n\n\n
      5. Ask for the source page.<\/strong> Include source_page so reviewers know where the date appeared in the PDF.<\/li>\n<\/ol>

        Here is a reusable prompt you can use for invoices, contracts, reports, and other business PDFs:<\/p>

        You are a date extraction assistant. From the attached PDF, extract every date that appears in the document.\nFor each date, return one JSON object with these fields:\n- date_iso: the date normalized to YYYY-MM-DD format\n- date_raw: the exact date text as written in the PDF\n- date_type: one of invoice_date, due_date, effective_date, expiration_date, issue_date, or other\n- confidence: a number from 0.0 to 1.0 showing how confident you are in the extracted value\n- hil_flag: true if confidence is below 0.85, the date format is ambiguous, or the date needs human review\n- source_page: the page number where the date appears\nRules:\n- Return a JSON array only.\n- Do not include commentary before or after the JSON.\n- Preserve the original date text in date_raw.\n- Use null for date_iso if the date cannot be normalized safely.\n- Mark ambiguous numeric dates, such as 03\/04\/2026, with hil_flag: true unless the document clearly indicates the locale.<\/pre>
        For example, OpenClaw should return output like this:<\/p>
        [\n{\n\"date_iso\": \"2026-03-14\",\n\"date_raw\": \"March 14, 2026\",\n\"date_type\": \"invoice_date\",\n\"confidence\": 0.98,\n\"hil_flag\": false,\n\"source_page\": 1\n},\n{\n\"date_iso\": \"2026-04-13\",\n\"date_raw\": \"04\/13\/2026\",\n\"date_type\": \"due_date\",\n\"confidence\": 0.86,\n\"hil_flag\": false,\n\"source_page\": 1\n},\n{\n\"date_iso\": \"2026-03-04\",\n\"date_raw\": \"03\/04\/2026\",\n\"date_type\": \"other\",\n\"confidence\": 0.72,\n\"hil_flag\": true,\n\"source_page\": 2\n}\n]<\/pre>
        The most important part of the prompt is the schema. Without a fixed schema, OpenClaw may return dates in different formats across different PDFs. One invoice might produce a paragraph, another might produce a list, and another might mix dates with unrelated document details. A fixed schema keeps every extraction result consistent.<\/p>
        You can also adjust the prompt for specific document types. For invoices, focus on invoice_date and due_date. For contracts, focus on effective_date and expiration_date. For reports, focus on issue_date, publication dates, reporting periods, and any date ranges that appear in the summary or cover page.<\/p>
        Use this invoice-specific version when you only need billing dates:<\/p>
        Extract the invoice_date and due_date from the attached PDF.\nReturn a JSON array only. Each object must include:\n- date_iso\n- date_raw\n- date_type\n- confidence\n- hil_flag\n- source_page\nUse invoice_date for the invoice issue date.\nUse due_date for the payment deadline.\nUse null for any missing field.\nFlag ambiguous or low-confidence dates with hil_flag: true.<\/pre>
        If the PDF comes from a known country or vendor, add a locale instruction to reduce date ambiguity:<\/p>
        This is a UK invoice. Interpret ambiguous numeric dates as DD\/MM\/YYYY unless the document clearly indicates another format.<\/pre>
        For US documents, use:<\/p>
        This is a US invoice. Interpret ambiguous numeric dates as MM\/DD\/YYYY unless the document clearly indicates another format.<\/pre>
        These small context instructions help OpenClaw handle dates like 03\/04\/2026, which can mean March 4 or 3 April depending on the document’s locale.<\/p>
        After the prompt returns structured date data, the next step is handling scanned PDFs where OpenClaw may need OCR before it can read the date text.<\/p>
        How do you extract dates from scanned PDFs with OCR?<\/h2>
        OpenClaw extracts dates from scanned PDFs by adding a searchable text layer with optical character recognition (OCR), then running the same date extraction prompt on the OCR-processed file. This extra step is necessary because scanned PDFs are image-based documents, so the PDF text may be empty, incomplete, or unreadable to the extraction workflow.<\/p>
        Use this OCR fallback when OpenClaw returns no useful text from the original PDF. For example, a scanned contract may look readable to a person, but OpenClaw may only see a page image unless OCR has converted the visible text into selectable text.<\/p>
        The basic scanned PDF workflow has four steps:<\/p>
          \n
        1. Run the PDF tool on the original file first. <\/strong>Start with the normal PDF extraction flow. Many business PDFs look like scans but still contain selectable text.<\/li>\n\n\n\n
        2. Check whether the extracted text is usable. <\/strong>Treat the file as scanned if the output is empty, mostly whitespace, garbled, or too short to contain the visible dates.<\/li>\n\n\n\n
        3. Run OCR on the PDF. <\/strong>Use an OCR tool such as ocrmypdf to create a new version of the file with a searchable text layer.<\/li>\n\n\n\n
        4. Run the date extraction prompt again. <\/strong>Send the OCR-processed PDF back to OpenClaw and use the same structured JSON prompt for date_iso, date_raw, date_type, confidence, hil_flag, and source_page.<\/li>\n<\/ol>
          Here is a simple OCR command your agent can run before retrying extraction:<\/p>
          ocrmypdf --skip-text input.pdf output-ocr.pdf<\/pre>
          The –skip-text flag tells OCRmyPDF to leave existing text alone and only OCR pages that need it. This is useful for mixed PDFs, where some pages are digital and others are scanned images.<\/p>
          After OCR finishes, run the OpenClaw PDF tool on the new file:<\/p>
          {\n\"tool\": \"pdf\",\n\"pdf\": \"workspace\/contracts\/vendor-contract-ocr.pdf\",\n\"prompt\": \"Extract every date in this PDF. Return a JSON array only with date_iso, date_raw, date_type, confidence, hil_flag, and source_page.\"\n}<\/pre>
          OCR improves scanned PDF extraction, but it does not guarantee perfect results. Scan quality affects date accuracy. Low-resolution scans, tilted pages, handwriting, stamps, watermarks, and poor contrast can cause OCR mistakes such as reading 03\/08\/2026 as 08\/08\/2026 or missing a date entirely.<\/p>
          For scanned PDFs, use stricter review rules than you would for born-digital PDFs. A good default is to set hil_flag: true when the confidence score is below 0.9, when the date is handwritten, or when the numeric format is ambiguous. This keeps uncertain dates out of your spreadsheet until a person checks them.<\/p>
          You can also add a scanned-document instruction to the prompt:<\/p>
          This PDF may contain OCR errors. Extract only dates that are clearly visible in the document. Preserve the exact OCR text in date_raw, normalize only when safe, and set hil_flag to true for unclear, handwritten, or ambiguous dates.<\/pre>
          If most of your files are scanned invoices, old contracts, receipts, or photographs, build OCR into the workflow rather than treating it as a one-time fix. In a self-managed OpenClaw setup, that usually means installing OCR tools and calling them before the PDF extraction step. In Managed OpenClaw, check whether your workflow can use a pre-built OCR skill or supported automation path before adding custom tools.<\/p>
          Once scanned PDFs are readable, the next optimization is limiting extraction to the pages where important dates usually appear.<\/p>
          How do you extract dates from specific PDF pages?<\/h2>
          Extract dates from specific PDF pages in OpenClaw by using page targeting when the relevant dates appear only in certain parts of the document. This is useful for long contracts, reports, policy documents, and agreements, where the important dates are usually listed on the cover page, signature page, summary page, or renewal schedule.<\/p>
          Page targeting helps the workflow in two ways. First, it reduces the amount of irrelevant text OpenClaw needs to process. Second, it lowers the chance that the model extracts unrelated dates from footers, revision histories, appendices, or legal references.<\/p>
          Use page targeting when you already know where the dates usually appear. For example:<\/p>
            \n
          • Invoices usually contain invoice_date and due_date on page 1.<\/li>\n\n\n\n
          • Contracts often contain effective_date near the first pages and signature dates near the last pages.<\/li>\n\n\n\n
          • Insurance policies may contain policy_start_date, policy_end_date, or renewal dates in the declarations page.<\/li>\n\n\n\n
          • Reports often contain issue_date, reporting period, or publication date on the cover page or executive summary.<\/li>\n<\/ul>
            Here is an example that extracts dates only from pages 1–3 and page 7:<\/p>
            {\n\"tool\": \"pdf\",\n\"pdf\": \"workspace\/contracts\/vendor-msa.pdf\",\n\"pages\": \"1-3,7\",\n\"prompt\": \"Extract every date from the selected PDF pages. Return a JSON array only with date_iso, date_raw, date_type, confidence, hil_flag, and source_page.\"\n}<\/pre>
            The pages value uses 1-based page numbers. Use a comma to separate individual pages and a dash to define a page range. For example, 1-3,7 means pages 1, 2, 3, and 7.<\/p>
            For contract workflows, you can make the prompt more specific:<\/p>
            Extract effective_date, expiration_date, renewal_date, and signature_date from the selected pages of this PDF.\nReturn a JSON array only. Each object must include:\n- date_iso\n- date_raw\n- date_type\n- confidence\n- hil_flag\n- source_page\nSet hil_flag to true if the date is ambiguous, missing context, or appears only in boilerplate legal text.<\/pre>
            Be careful with long PDFs because not every date is equally important. A 50-page contract may include dozens of dates in examples, change logs, references, or attachment labels. Page targeting helps OpenClaw focus on the pages where business-critical dates are most likely to appear.<\/p>
            If you are not sure which pages contain the relevant dates, use a two-pass workflow:<\/p>
              \n
            1. Scan the document structure first.<\/strong> Ask OpenClaw to identify pages that contain date-heavy sections, such as cover pages, tables, signature blocks, renewal terms, or invoice headers.<\/li>\n\n\n\n
            2. Run date extraction only on those pages.<\/strong> Use the selected page numbers in the pages field and apply the structured date extraction prompt.<\/li>\n<\/ol>
              For example, the first pass can use this prompt:<\/p>
              Review this PDF and identify which pages are most likely to contain business-critical dates, such as invoice dates, due dates, effective dates, expiration dates, renewal dates, issue dates, or signature dates. Return only a JSON array of page numbers with a short reason for each page.<\/pre>
              Then the second pass extracts dates from those selected pages.<\/p>
              For scanned PDFs, run OCR before page-targeted extraction if the selected pages are image-based. OCR should occur before the final extraction step so OpenClaw can accurately read the date text from the targeted pages.<\/p>
              Page targeting works best when documents follow a consistent layout. If every vendor invoice places the due date on page 1, or every contract template places the term dates in the first three pages, you can hard-code those page ranges into the workflow. If layouts vary, use the two-pass approach, so OpenClaw finds the likely pages before extracting the final date values.<\/p>
              After extracting dates from the correct pages, validate the output before exporting. Page targeting reduces noise, but validation still catches ambiguous formats, impossible dates, and low-confidence values.<\/p>
              Should you use native PDF mode or fallback mode for date extraction?<\/h2>
              Use native PDF mode<\/strong> for the most direct way to read short, layout-sensitive PDFs, and fallback mode<\/strong> for<\/span> page targeting, custom processing, or support for a wider range of models. Both modes can extract dates from PDFs, but they process the document differently.<\/p>
              Native PDF mode sends the PDF directly to a model that supports PDF input. This works well for invoices, contracts, receipts, and reports where the model needs to understand the layout around each date. For example, native mode can use nearby labels like “Invoice date,” “Payment due,” “Effective date,” or “Expiration date” to classify each extracted value more accurately.<\/p>
              Fallback mode extracts the PDF content first, then sends the extracted text or page content to the selected model. This mode is useful when you want to target specific pages, use a model that does not support native PDF input, or add extra preprocessing steps before extraction.<\/p>
              For most date extraction workflows, start with this rule:<\/p>
              Use native PDF mode for short PDFs where layout matters.\nUse fallback mode for long PDFs, page targeting, or custom preprocessing.<\/pre>
              Native PDF mode is usually the better starting point for documents like:<\/p>
                \n
              • One-page invoices with invoice_date and due_date.<\/li>\n\n\n\n
              • Short contracts with effective_date and expiration_date.<\/li>\n\n\n\n
              • Receipts or statements where dates appear close to labels, totals, or payment terms.<\/li>\n\n\n\n
              • PDFs where table structure or visual layout helps explain what each date means.<\/li>\n<\/ul>
                Fallback mode is usually better for documents like:<\/p>
                  \n
                • Long reports where dates only appear on selected pages.<\/li>\n\n\n\n
                • Contracts where only the first pages and signature pages matter.<\/li>\n\n\n\n
                • Files that need OCR, cleanup, or text extraction before analysis.<\/li>\n\n\n\n
                • Workflows that need the pages field to reduce irrelevant content.<\/li>\n\n\n\n
                • Pipelines that rely on a specific model or automation step.<\/li>\n<\/ul>
                  Here is a simple comparison:<\/p>
                  For example, use native mode for a two-page supplier invoice because the model can read the invoice header, due date label, payment terms, and table layout together. This context helps OpenClaw distinguish the invoice_date from the due_date.<\/p>
                  Use fallback mode for a 50-page master service agreement, with the relevant dates appearing on pages 1–3 and the signature page. In this case, processing the whole PDF may introduce noise from boilerplate dates, amendment references, or appendix examples. Fallback mode with page targeting keeps the extraction focused on the pages most likely to contain business-critical dates.<\/p>
                  If accuracy is the priority, test both modes on a small sample of your real PDFs before choosing one for the full workflow. Run 10–20 documents through each mode, compare the extracted date_iso, date_type, confidence, and source_page, and choose the mode that produces the fewest human review flags.<\/p>
                  A practical setup can use both modes:<\/p>
                    \n
                  1. Try native PDF mode first<\/strong> for short, clean PDFs.<\/li>\n\n\n\n
                  2. Switch to fallback mode<\/strong> when the document is long or needs page targeting.<\/li>\n\n\n\n
                  3. Run OCR before extraction<\/strong> when the PDF is scanned or image-based.<\/li>\n\n\n\n
                  4. Validate all extracted dates<\/strong> before sending them to your spreadsheet or database.<\/li>\n<\/ol>
                    The mode only controls how OpenClaw reads the PDF. The final reliability still depends on your prompt, date schema, OCR quality, page selection, and validation rules. After choosing the mode, the next step is checking whether the extracted dates are valid enough to use automatically.<\/p>
                    How do you validate extracted dates before exporting them?<\/h2>
                    Validate extracted dates by checking their format, meaning, confidence score, and business logic before sending them to a spreadsheet, CRM, accounting tool, or database. This step prevents incorrect dates from automatically entering your system, especially when PDFs contain ambiguous formats, OCR errors, repeated boilerplate dates, or dates that do not belong in the field you need.<\/p>
                    A reliable OpenClaw date extraction workflow should validate every result with four rules:<\/p>
                      \n
                    1. Normalize every date to ISO 8601 format.<\/strong> Store dates as YYYY-MM-DD before export. For example, March 14, 2026, 14\/03\/2026, and 2026-03-14 should all be treated as 2026-03-14 when the meaning is clear. Keep the original PDF text in date_raw so you can audit how OpenClaw interpreted the value.<\/li>\n\n\n\n
                    2. Flag ambiguous date formats.<\/strong> Dates like 03\/04\/2026 can mean March 4 or 3 April depending on the document’s locale. Use clues such as currency, address, language, vendor country, and tax format to decide whether the document follows MM\/DD\/YYYY or DD\/MM\/YYYY. If the workflow cannot confidently detect the locale, set hil_flag to true and send the row to human review.<\/li>\n\n\n\n
                    3. Apply sanity bounds.<\/strong> Check whether each date makes sense for the document type. For invoices, the invoice_date should usually be recent, and the due_date should not come before the invoice_date. For contracts, the expiration_date should come after the effective_date. For reports, the issue_date should not be years outside the reporting period unless the document clearly explains why.<\/li>\n\n\n\n
                    4. Route low-confidence values to human review.<\/strong> Use the confidence score and hil_flag field to separate reliable dates from uncertain dates. A practical default is to flag any value below 0.85 for born-digital PDFs and below 0.9 for scanned or OCR-processed PDFs.<\/li>\n<\/ol>
                      Here is an example of a validated date object:<\/p>
                      {\n\"date_iso\": \"2026-03-04\",\n\"date_raw\": \"03\/04\/2026\",\n\"date_type\": \"due_date\",\n\"confidence\": 0.72,\n\"hil_flag\": true,\n\"source_page\": 1,\n\"validation_reason\": \"Ambiguous numeric date format\"\n}<\/pre>
                      Validation should happen after OpenClaw returns the structured JSON and before the data reaches the final destination. The workflow can reject invalid dates, flag uncertain dates, or add a validation_reason field that explains why a human should review the value.<\/p>
                      For invoices, use validation rules like these:<\/p>
                      If invoice_date is missing, flag the record.\nIf due_date is missing, flag the record.\nIf due_date is earlier than invoice_date, flag the record.\nIf invoice_date is in the future, flag the record.\nIf invoice_date is older than the allowed accounting window, flag the record.\nIf confidence is below 0.85, flag the record.\nIf date_raw uses an ambiguous numeric format, flag the record.<\/pre>
                      For contracts, use different rules:<\/p>
                      If effective_date is missing, flag the record.\nIf expiration_date is missing, flag the record.\nIf expiration_date is earlier than effective_date, flag the record.\nIf renewal_date appears without renewal terms, flag the record.\nIf signature_date appears after effective_date, flag the record for review.\nIf confidence is below 0.85, flag the record.<\/pre>
                      You can also ask OpenClaw to include validation signals directly in the extraction output:<\/p>
                      After extracting each date, validate it against the document context.\nReturn a JSON array only. Each object must include:\n- date_iso\n- date_raw\n- date_type\n- confidence\n- hil_flag\n- source_page\n- validation_reason\nSet hil_flag to true if:\n- the date format is ambiguous\n- the confidence score is below 0.85\n- the date conflicts with another extracted date\n- the date appears in boilerplate text, an example, or a revision history\n- the date cannot be normalized safely<\/pre>
                      For automated exports, keep two output paths. Send clean records directly to the final spreadsheet or database, and send flagged records to a review queue. The review queue can be a separate sheet, Airtable view, ticket, or JSON file that only contains rows where hil_flag is true.<\/p>
                      This split keeps the workflow efficient. High-confidence dates are automatically moved, while uncertain dates are still checked before they affect payment deadlines, contract renewals, compliance reports, or customer records.<\/p>
                      After the validation rules are in place, you can decide whether to keep using custom prompts or replace part of the workflow with a pre-built OpenClaw skill.<\/p>
                      Can OpenClaw skills extract dates from PDFs automatically?<\/h2>
                      OpenClaw skills can help extract dates from PDFs automatically when the task follows a repeatable workflow, such as reading invoices, contracts, receipts, or reports and returning the same date fields each time. A skill is not a replacement for the PDF tool or your extraction prompt. It is a reusable instruction package that specifies when to use specific tools, how to process the file, and how to format the result.<\/p>
                      In OpenClaw, skills are stored as folders that include a SKILL.md file with instructions and configuration details. OpenClaw loads bundled skills and optional workspace skills based on the agent’s environment, configuration, and available dependencies. ClawHub also works as a public registry for discovering and installing OpenClaw skill bundles. <\/p>
                      Use a skill when the PDF date extraction process is stable. For example, an invoice date extraction skill could tell OpenClaw to check the invoice header, find invoice_date and due_date, normalize each value to ISO 8601, return JSON, and flag ambiguous dates for review. This avoids having to rewrite the same prompt every time you process a new batch of invoices.<\/p>
                      A date extraction skill should define:<\/p>
                        \n
                      1. When the skill should run. <\/strong>For example, run it when the user requests the extraction of dates from invoice PDFs, contract PDFs, renewal documents, or scanned PDF files.<\/li>\n\n\n\n
                      2. Which tools the agent should use. <\/strong>The skill can instruct OpenClaw to use the PDF tool first, run OCR when the file is scanned, and apply validation before export.<\/li>\n\n\n\n
                      3. Which date fields to return. <\/strong>Common fields include invoice_date, due_date, effective_date, expiration_date, renewal_date, issue_date, signature_date, and other.<\/li>\n\n\n\n
                      4. Which output schema to follow. <\/strong>The skill should return consistent fields such as date_iso, date_raw, date_type, confidence, hil_flag, source_page, and validation_reason.<\/li>\n\n\n\n
                      5. Which values need human review. <\/strong>The skill should set hil_flag: true for low-confidence dates, ambiguous numeric formats, OCR errors, missing context, or dates that fail business rules.<\/li>\n<\/ol>
                        Here is a simple example of what the instruction inside a date extraction skill could say:<\/p>
                        Use this skill when the user asks to extract dates from PDF invoices, contracts, reports, or scanned PDF documents.\nWorkflow:\n1. Read the PDF with the OpenClaw PDF tool.\n2. If the PDF text is empty or unreadable, run OCR before extraction.\n3. Extract all business-critical dates.\n4. Return a JSON array only.\n5. Include date_iso, date_raw, date_type, confidence, hil_flag, source_page, and validation_reason.\n6. Set hil_flag to true for ambiguous, low-confidence, handwritten, or contextless dates.\n7. Do not export flagged dates without human review.<\/pre>
                        For simple one-off tasks, a prompt is usually enough. Use the reusable prompt when you only need to extract dates from a few PDFs or when you are still testing which fields matter. A prompt is easier to edit and debug.<\/p>
                        Use a skill when the workflow becomes repetitive. For example, a finance team that processes supplier invoices every night can use a skill to apply the same extraction schema, validation rules, and export behavior across every file. A legal team can use a different skill set for contract dates, since contracts require fields such as effective_date, expiration_date, renewal_date, and signature_date, rather than invoice-specific fields.<\/p>
                        You can also install PDF-focused skills from ClawHub or create your own workspace skill. For example, ClawHub lists PDF extraction skills that process PDF content for downstream model workflows, and OpenClaw’s CLI supports searching and installing skills from ClawHub.<\/p>
                        Before using a third-party skill, inspect its SKILL.md, dependencies, install commands, and requested permissions. Treat community skills as untrusted until reviewed, especially if they ask for shell access, file-system access, API keys, or network permissions. This matters for PDF data extraction because invoices, contracts, and reports often contain sensitive business data.<\/p>
                        A practical rule is:<\/p>
                        Use a prompt for testing or one-off PDF date extraction.\nUse a skill for repeatable invoice, contract, report, or batch extraction workflows.\nUse a custom skill when your date fields, validation rules, or export destination are specific to your business.<\/pre>
                        After deciding between a prompt and a skill, the next step is to turn the workflow into a scheduled batch process so OpenClaw can automatically extract dates from new PDFs.<\/p>
                        How do you schedule batch PDF date extraction in OpenClaw?<\/h2>
                        You schedule batch PDF date extraction in OpenClaw by creating a repeatable workflow that watches a folder, processes new PDFs, validates the extracted dates, and writes the clean output to a spreadsheet, database, or results folder. This turns PDF date extraction from a manual task into an automated pipeline for invoices, contracts, reports, and renewal documents.<\/p>
                        A basic batch workflow has five parts:<\/p>
                          \n
                        1. A watch folder for incoming PDFs. <\/strong>Create a dedicated folder for files that need date extraction, such as workspace\/extraction-queue\/. Add new invoices, contracts, or reports to this folder by upload, email forwarding, scanner export, or another automation.<\/li>\n\n\n\n
                        2. A date extraction task. <\/strong>Configure the task to read each new PDF, run the OpenClaw PDF tool, and apply your date extraction prompt or skill. The task should return structured fields such as date_iso, date_raw, date_type, confidence, hil_flag, and source_page.<\/li>\n\n\n\n
                        3. A validation task. <\/strong>Run validation after extraction to check date format, confidence score, date order, and business rules. For example, invoices should not have a due_date earlier than the invoice_date, and contracts should not have an expiration_date earlier than the effective_date.<\/li>\n\n\n\n
                        4. An output location. <\/strong>Save clean results to a folder such as workspace\/extraction-results\/, or send them to Google Sheets, Airtable, a CRM, an accounting platform, or a contract tracker.<\/li>\n\n\n\n
                        5. A review queue for flagged dates. <\/strong>Send records with hil_flag: true to a separate file, sheet, or task queue so a person can review ambiguous dates before they enter the final system.<\/li>\n<\/ol>
                          For example, your workspace can use this folder structure:<\/p>
                          workspace\/\nextraction-queue\/\ninvoice-001.pdf\ncontract-042.pdf\nextraction-results\/\nclean-dates.json\nflagged-dates.json\nprocessed\/\ninvoice-001.pdf\ncontract-042.pdf<\/pre>
                          The scheduled workflow should process files in small batches instead of sending every PDF at once. This avoids file-count limits, reduces failures, and makes retries easier. For example, the task can process up to 10 PDFs, write the result, move completed files to processed\/, and leave failed files in the queue with an error note.<\/p>
                          A simple batch instruction can look like this:<\/p>
                          Every scheduled run:\n1. List new PDF files in workspace\/extraction-queue\/.\n2. Process up to 10 PDFs at a time.\n3. Run the date extraction prompt or date extraction skill.\n4. Save high-confidence results to workspace\/extraction-results\/clean-dates.json.\n5. Save low-confidence or ambiguous results to workspace\/extraction-results\/flagged-dates.json.\n6. Move successfully processed PDFs to workspace\/processed\/.\n7. Leave failed PDFs in the queue and write the error reason.<\/pre>
                          If OpenClaw uses a heartbeat or scheduled task file in your setup, define separate runs for extraction and validation. This keeps the workflow easier to debug because extraction errors and validation errors are handled separately.<\/p>
                          # Example schedule\n0 1 * * *  extract_dates_from_queue\n30 1 * * * validate_extracted_dates\n45 1 * * * export_clean_dates<\/pre>
                          In this example, OpenClaw starts extracting dates at 1:00 AM, validates the results at 1:30 AM, and exports clean records at 1:45 AM. The exact schedule can change based on your document volume, API limits, and how quickly downstream systems need the data.<\/p>
                          For overnight or recurring extraction, the hosting environment matters. A local OpenClaw instance only runs while the computer is on and connected. If the laptop sleeps, shuts down, or loses internet access, the scheduled job stops. Managed OpenClaw by Hostinger is a natural fit for this type of workflow because it gives you a 1-click OpenClaw environment that stays online without requiring you to maintain the server yourself.<\/p>
                          When you want a scheduled PDF date-extraction workflow for business documents but do not need root-level customization. Use a self-managed VPS setup when the workflow requires custom OCR packages, private network access, modified skills, or deeper control over the runtime environment.<\/p>
                          For larger queues, add two safeguards:<\/p>