meta-document-to-decision

star 4.3k

Use this meta-skill instead of answering directly when the current user provides or references a document, contract, quote, spreadsheet, notice, or paperwork and asks for a decision-ready analysis: sign/reject/negotiate, renewal risk, evidence table, questions to ask, or concrete next action. It may inspect PDF/DOCX/XLSX/pasted excerpts. Do not use it for generic summarization, generic report writing, standalone sales emails, generic contract-term explanations, or document text that is merely quoted as historical context.

opensquilla By opensquilla schedule Updated 6/12/2026
play_arrow Run Skill in Manus View GitHub

name: meta-document-to-decision description: "Use this meta-skill instead of answering directly when the current user provides or references a document, contract, quote, spreadsheet, notice, or paperwork and asks for a decision-ready analysis: sign/reject/negotiate, renewal risk, evidence table, questions to ask, or concrete next action. It may inspect PDF/DOCX/XLSX/pasted excerpts. Do not use it for generic summarization, generic report writing, standalone sales emails, generic contract-term explanations, or document text that is merely quoted as historical context." kind: meta meta_priority: 67 always: false final_text_mode: "step:decision_brief_audit" triggers:

  • "document decision"
  • "vendor renewal"
  • "analyze renewal materials"
  • "analyze this contract excerpt for a decision"
  • "decide whether to sign this document"
  • "decide tomorrow whether to sign"
  • "sign, reject, or negotiate"
  • "evidence table for this document"
  • "questions for the vendor about this contract"
  • "看下这个文件再决定"
  • "帮我判断这个文档"
  • "合同风险"
  • "报价单分析"
  • "文件里我该注意什么"
  • "读完这份合同告诉我签不签"
  • "供应商续费"
  • "续费材料"
  • "这份报价单要不要接受"
  • "合同自动续约风险"
  • "合同付款期限风险"
  • "这个合同要不要签" provenance: origin: opensquilla-original license: Apache-2.0 metadata: opensquilla: risk: medium capabilities: [filesystem-read, filesystem-write] clawhub_top100_composition:
    • skill: "Word / DOCX" local_skill: docx rank_source: "Top ClawHub Skills downloads top100, 2026-05-28" rank: 28 role: "Inspect Word contracts, notices, and decision documents."
    • skill: "Excel / XLSX" local_skill: xlsx rank_source: "Top ClawHub Skills downloads top100, 2026-05-28" rank: 31 role: "Inspect spreadsheet quotes, totals, dates, and formula outputs."
    • skill: "Pdf" local_skill: pdf-toolkit rank_source: "Top ClawHub Skills downloads top100, 2026-05-28" rank: 36 role: "Extract PDF text, tables, titles, and page references."

composition: steps: - id: intake kind: llm_chat with: system: "You classify document decision requests and preserve every path, URL, excerpt, and decision question." task: | Parse the request into a decision-analysis contract.

      Request:
      {{ inputs.user_message | xml_escape | truncate(4000) }}

      Return exactly:
      DOCUMENT_TYPES:
        - <pdf|docx|xlsx|pasted_text|unknown>
      SOURCES:
        - <path/url/excerpt label>
      DECISION_QUESTION: <question or ASSUMED: what should the user do next>
      RISK_DOMAIN: <contract|finance|school|medical|operations|other>
      NEEDS_CLARIFICATION: <yes|no>
      MISSING_FIELDS:
        - <source_material|decision_question|none>
      OUTPUT_LANGUAGE: <language>
      Set NEEDS_CLARIFICATION: no when SOURCES has at least one path, URL,
      or pasted excerpt and DECISION_QUESTION is explicit or can be safely
      assumed from the user's ask. In that case MISSING_FIELDS must be
      exactly "- none".
- id: clarify
  kind: user_input
  depends_on: [intake]
  when: "'NEEDS_CLARIFICATION: yes' in outputs.intake and '- none' not in outputs.intake"
  clarify:
    mode: form
    intro: "文档决策分析缺少材料或问题。请补齐后我会继续。"
    nl_extract: true
    fields:
      - name: source_material
        type: string
        required: true
        prompt: "文件路径、URL 或摘录 / Source material"
        max_chars: 3000
      - name: decision_question
        type: string
        prompt: "你要做的决定 / Decision question"
        max_chars: 300
    cancel_keywords: ["取消", "算了", "cancel", "stop"]
    timeout_hours: 24
- id: pdf_extract
  kind: skill_exec
  skill: pdf-toolkit
  depends_on: [intake, clarify]
  when: "'pdf' in (outputs.intake | lower)"
  on_failure: pdf_extract_fallback
  with:
    task: "Extract text, tables, document title, and page references for this decision analysis: {{ outputs.intake | truncate(1200) }}"
- id: docx_extract
  kind: skill_exec
  skill: docx
  depends_on: [intake, clarify]
  when: "'docx' in (outputs.intake | lower) or 'word' in (outputs.intake | lower)"
  on_failure: docx_extract_fallback
  with:
    task: "Inspect document text, headings, tracked-change hints, tables, and clauses for this decision analysis."
- id: xlsx_extract
  kind: skill_exec
  skill: xlsx
  depends_on: [intake, clarify]
  when: "'xlsx' in (outputs.intake | lower) or 'spreadsheet' in (outputs.intake | lower)"
  on_failure: xlsx_extract_fallback
  with:
    task: "Inspect sheets, tables, totals, formula outputs, and anomalies for this decision analysis."
- id: pasted_text_extract
  kind: llm_chat
  depends_on: [intake, clarify]
  when: "'pasted_text' in (outputs.intake | lower) or 'unknown' in (outputs.intake | lower)"
  with:
    system: "You turn pasted document excerpts into a source-labeled evidence packet without inventing missing clauses."
    task: |
      Build an evidence packet from the user's pasted materials only.
      Preserve source labels such as quote, contract excerpt, email,
      notice, spreadsheet excerpt, or unknown excerpt. Extract exact
      money amounts, dates, obligations, contradictions, and missing facts.

      Intake:
      {{ outputs.intake | truncate(1200) }}

      Request:
      {{ inputs.user_message | xml_escape | truncate(6000) }}
- id: pdf_extract_fallback
  kind: llm_chat
  with:
    system: "You build a limited PDF evidence packet from only the user's pasted text and explicit file names."
    task: |
      Return a PDF evidence packet. Mark unavailable file extraction clearly.

      Request:
      {{ inputs.user_message | xml_escape | truncate(5000) }}
- id: docx_extract_fallback
  kind: llm_chat
  with:
    system: "You build a limited DOCX evidence packet from only the user's pasted text and explicit file names."
    task: |
      Return a DOCX evidence packet. Mark unavailable file extraction clearly.

      Request:
      {{ inputs.user_message | xml_escape | truncate(5000) }}
- id: xlsx_extract_fallback
  kind: llm_chat
  with:
    system: "You build a limited spreadsheet evidence packet from only the user's pasted text and explicit file names."
    task: |
      Return a spreadsheet evidence packet. Mark unavailable file extraction clearly.

      Request:
      {{ inputs.user_message | xml_escape | truncate(5000) }}
- id: risk_review
  kind: llm_chat
  depends_on: [pdf_extract, docx_extract, xlsx_extract, pasted_text_extract]
  with:
    system: "You identify document risks and decision-relevant evidence without giving regulated professional advice."
    task: |
      Extract:
      - key facts with evidence source/page/sheet when available
      - money/date/obligation/risk clauses
      - inconsistencies and missing information
      - decisions the user can safely make
      - items requiring lawyer/doctor/accountant/professional review
      Compare dates against the current runtime date when available. Do
      not mark a payment deadline, cancellation window, or event as
      overdue unless the date is actually before the current date; when
      uncertain, call it "upcoming/待确认" instead of expired. Do not
      infer that a cancellation window has passed from a payment due date
      alone; require an explicit contract end, renewal effective date, or
      cancellation deadline.
      Do not derive cancellation deadlines by subtracting days from invoice or payment due dates.
      If the contract end date or renewal effective date is missing, mark the
      cancellation deadline unknown / 待确认 and avoid saying the notice window has passed.

      Intake:
      {{ outputs.intake | truncate(1200) }}
      PDF:
      {{ outputs.pdf_extract | truncate(3000) }}
      DOCX:
      {{ outputs.docx_extract | truncate(3000) }}
      XLSX:
      {{ outputs.xlsx_extract | truncate(3000) }}
      Pasted text:
      {{ outputs.pasted_text_extract | truncate(4000) }}
- id: decision_brief
  kind: llm_chat
  depends_on: [risk_review]
  with:
    system: "You write decision briefs for ordinary users and managers."
    task: |
      Return a decision-ready brief:
      1. Bottom-line recommendation / 底线推荐
         Start with a one-paragraph boss-forwardable summary that states
         sign / negotiate first / reject, the decisive reason, money/date
         exposure, and the next owner/action.
      2. Evidence table. Use the literal section title "Evidence table /
         证据表" and cite each row's source as quote/contract/email/page/
         sheet/excerpt when available.
      3. Risks ranked high/medium/low
      4. Questions to ask the other party
      5. What to do next in 24 hours
      6. Professional-review caveats where needed. For contract,
         finance, medical, school, or regulated decisions, include an
         explicit "Professional-review caveat / 专业复核提醒" section
         naming the right reviewer, such as lawyer/律师, accountant/会计,
         doctor/医生, school administrator, or compliance owner. Do not
         bury this inside the risk table.
      Do not claim to create, save, export, download, or attach a file
      unless an explicit export step ran. Return the usable brief inline.
      Preserve date status accurately: if a deadline is after the current
      date, describe it as upcoming, not overdue.
      Do not say a cancellation window has already passed unless the
      evidence includes the contract end date, renewal effective date, or
      cancellation deadline.
      If the only known future date is a payment deadline, do not compute
      a cancellation deadline from that date. Say the cancellation window
      is "unknown / 待确认" and ask the supplier to confirm whether the
      30-day notice period is measured from contract end, renewal start,
      invoice due date, or another date.
      Do not derive cancellation deadlines by subtracting days from invoice or payment due dates.
      If the contract end date or renewal effective date is missing, keep the
      cancellation deadline unknown and avoid saying the notice window has passed.
      Do not speculate that the notice period may already be too short
      unless the evidence contains the date from which the notice period
      runs. Phrase unknown notice status as a negotiation question, not as
      a likely risk conclusion.
      Keep the brief boss-forwardable: no workflow commentary, no meta-skill
      names, no private reasoning, and no broad legal lecture. Do not invent
      exact reply deadlines such as "16:00" unless the user provided that
      time; use natural windows like today, tomorrow morning, or before the
      payment due date. Do not cite statutes or legal article numbers unless
      they appear in the provided materials. Prefer practical negotiation
      language over categorical legal conclusions.
      Preserve the user's language. If the original request is in English,
      write the final brief in English with English section headings only;
      do not default to Chinese or bilingual headings. If the original
      request is in Chinese, write Simplified Chinese and do not default to
      English headings.

      Risk review:
      {{ outputs.risk_review | truncate(7000) }}
- id: decision_brief_audit
  kind: llm_chat
  depends_on: [decision_brief, risk_review]
  with:
    system: "You are the final quality gate for a document-to-decision brief. Return only the cleaned final answer the user should read. Do not explain the audit. Keep it boss-forwardable and source-faithful."
    task: |
      Clean the draft decision brief below.

      Draft brief:
      {{ outputs.get('decision_brief', '') | truncate(9000) }}

      Risk review:
      {{ outputs.get('risk_review', '') | truncate(5000) }}

      Original request:
      {{ inputs.user_message | xml_escape | truncate(5000) }}

      Required audit rules:
      - Remove statutes, legal article numbers, and law-name citations
        such as Civil Code, 民法典, article numbers, court standards, or
        specific legal doctrines unless they appear in the provided
        materials. Replace them with "ask counsel to review enforceability".
      - Do not invent today's date or the current calendar date. If the
        runtime date is not explicitly present in the source material, use
        relative wording such as today or before the payment due date.
      - If the source says sales email, do not call it oral or verbal.
        Treat it as a written sales email that conflicts with the contract
        excerpt, not as a spoken promise.
      - Do not say an email promise is legally invalid, almost invalid, or
        unenforceable as a categorical conclusion. Say it may be weaker
        than a signed contract amendment and should be confirmed in a
        contract addendum or signed written clarification.
      - Remove categorical Chinese legal phrases such as 效力弱于合同 or
        书面条款优先 unless the provided material says so. Prefer "销售邮件
        与合同摘录冲突，需要供应商用补充协议或签字书面澄清确认".
      - Preserve the user's requested structure: recommendation, evidence
        table, high/medium/low risks, supplier questions, next 24 hours,
        and professional-review caveat.
      - Preserve the original request language. For an English request,
        return an English-only brief with English headings; remove Chinese
        headings such as 底线推荐, 证据表, 高中低风险, 要问供应商的问题, 接下来
        24 小时, and 专业复核提醒 unless they are quoted source text. For a
        Chinese request, return Simplified Chinese and do not switch to
        English section headings.
      - Use these exact section titles for Chinese or bilingual requests
        so the brief is easy to scan:
        "Bottom-line recommendation / 底线推荐",
        "Evidence table / 证据表",
        "Risks ranked high/medium/low / 高中低风险",
        "Questions to ask the supplier / 要问供应商的问题",
        "What to do next in 24 hours / 接下来 24 小时",
        and "Professional-review caveat / 专业复核提醒".
        For English-only requests, use the English portion of each heading
        only: Bottom-line recommendation, Evidence table, Risks ranked
        high/medium/low, Questions to ask the supplier, What to do next in
        24 hours, Professional-review caveat.
      - Preserve date status accurately: payment due 2026-06-03 is an
        upcoming payment deadline, not overdue. Do not infer any
        cancellation deadline from that payment due date.
      - Keep the answer concise and directly forwardable to a boss. Remove
        workflow commentary, meta-skill names, runtime details, and private
        reasoning.

Document To Decision

Converts mixed business and life documents into evidence-backed decision briefs.

Install via CLI
npx skills add https://github.com/opensquilla/opensquilla --skill meta-document-to-decision
Repository Details
star Stars 4,280
call_split Forks 335
navigation Branch main
article Path SKILL.md
Occupations
Management Analysts 131111
More from Creator
opensquilla
opensquilla Explore all skills →