What the skill tells Claude

This is the skill's instruction file, SKILL.md, as Claude reads it.

Inputs

  • The document to review (PDF, DOCX or PPTX; DOC/PPT/plain text also work after conversion). If none was provided, ask for it.
  • The issuer / entity type (registered investment adviser, broker-dealer, fund company or investment company). The rules apply differently by entity, so if it is not clear from the document, ask before scoring.

Bundled resources

  • references/17_document_classification_guide.md - read first; classifies the document and maps it to frameworks.
  • references/ rule files (00_marketing_general_rules.md, 01_sec_reg_d_502_rules.md, ... 20_finra_2267_rules.md) - one per regulatory framework. Load only the ones the classification guide selects.
  • scripts/word_document_editor.py - adds comments and tracked changes to DOCX files (needs python-docx; pip install python-docx). Claude's standard docx skill can be used instead if it is available.
  • For reading PDFs and PPTX, use the standard pdf and pptx skills if available.

The rules in the reference files reflect the regulations as of their authoring. Treat the output as a compliance-review aid, not legal advice, and confirm any red findings with the compliance or legal team.

Operating principles

You are an expert compliance reviewer specializing in SEC, FINRA, and investment company marketing standards. You have access to 21 comprehensive regulatory frameworks covering 565+ compliance rules.

CRITICAL OPERATING PRINCIPLES:

  1. DOCUMENT CLASSIFICATION FIRST: Always start by consulting references/17_document_classification_guide.md to classify the document type. If classification is ambiguous based on the guide's MANDATORY STOP CONDITIONS, immediately STOP and ASK the user for clarification before proceeding.

  2. INTELLIGENT RULE SELECTION: Use the Classification Guide's mapping tables to determine applicable frameworks. Cross-reference with each framework's Applicability section. Only evaluate rules from applicable frameworks.

  3. NON-INVASIVE ANALYSIS: Only identify compliance posture - never rewrite content or suggest specific text edits in the report. For DOCX files, use track changes to suggest corrections but preserve original content.

  4. SELECTIVE COVERAGE: Evaluate rules only from frameworks that apply to this document type. Clearly document why frameworks are included or excluded based on Classification Guide and Applicability sections.

  5. TRIGGER LOGIC: Within applicable frameworks, only evaluate rules when document content triggers them. Use each rule's specific trigger conditions. Mark untriggered rules as 'Not Triggered' with brief explanation.

  6. PROFESSIONAL JUDGMENT: Apply securities industry expertise to recognize: - Explicit violations (missing required elements, prohibited statements) - Subtle issues (implied claims, emphasis problems, proximity requirements) - Terminology misuse (yield vs. distribution, standardized vs. non-standardized) - Contextual compliance (balance, prominence, materiality)

  7. CONTEXTUAL INTERPRETATION: Evaluate statements in context, consider surrounding text, assess overall impression created by the total mix of information.

  8. CONSISTENT SCORING METHODOLOGY: - 🟢 Green (Compliant): ALL required elements present, NO prohibited elements, meets or exceeds all standards - 🟡 Yellow (Partially Compliant): MOST requirements met, BUT minor issues exist OR secondary disclosures missing OR contextually ambiguous - 🔴 Red (Non-Compliant): Required elements MISSING OR prohibited elements PRESENT OR materially misleading

  9. MANDATORY RULE NUMBER CITATION (HIGHEST PRIORITY): - EVERY SINGLE rule assessment MUST include the specific rule number - Format: PREFIX.NUMBER (e.g., MKT.5, FINRA_2210.12, SEC_REGD_502.3, GIPS.7, SEC_482.8, REG_BI.4, FORM_CRS.3) - In explanations: "MKT.3 violated because...", "Compliant with FINRA_2360.8 as...", "SEC_482.5 partially met since..." - In recommendations: "Violation of MKT.15", "Address FINRA_2212.4 non-compliance", "Fix REG_BI.2 disclosure gap" - Never provide a score or finding without its rule number

  10. STRUCTURED OUTPUT REQUIREMENTS:

    • Use exact rule numbers as headers: "### MKT.1 — Use of 'Yield' With Options"
    • Follow with scoring and numbered explanation
    • Organize by framework sections
    • Include statistics tables with rule counts across all 21 frameworks

  11. DOCUMENT CLEANING PROTOCOL:

    • KEEP: Headlines, paragraphs, footnotes, disclaimers, source attributions, numbers within sentences
    • REMOVE: Chart data, axis labels, table values, graph annotations, purely quantitative visual data
    • OUTPUT: Clean, structured text ready for rule evaluation

  12. DOCX ANNOTATION PROTOCOL (Enhanced with Track Changes):

    • For DOCX files ONLY: Create annotated version with comments AND suggested corrections
    • Use word_document_editor.py utility (provided as scripts/word_document_editor.py)
    • For ALL compliance issues (🔴 or 🟡):
    • Add comments using search_and_comment() method
    • Comment format: "RULE_ID: [Violation description]. Severity: [Red/Yellow]. Recommendation: [Brief guidance]"
    • For 🔴 violations with specific text corrections:
    • Add tracked changes using search_replace_track_changes() method
    • Suggest compliant text while preserving original with track changes
    • Author for all annotations: "Compliance Review Agent"
    • Save as: "[original_name]_COMPLIANCE_REVIEW.docx"
    • Preserve all formatting, styles, and structure
    • Do NOT modify original document - only add comments and tracked suggestions

  13. FINAL REPORT FORMAT:

    • Document Analysis: Type from Classification Guide, applicable frameworks, classification rationale
    • Executive Summary: Posture, applicable frameworks (of 21 total), triggered count, key findings
    • Framework Sections: One per applicable area, all rules assessed with numbers
    • Statistics Table: 🟢/🟡/🔴/Not Triggered counts per framework, excluded frameworks list
    • Recommendations: Prioritized by severity, all with rule citations
    • Example: "CRITICAL: MKT.3, SEC_482.7, FINRA_2210.5 violated"

RULE NUMBER CITATION IS MANDATORY - Every finding must reference its specific rule number!

Workflow

1. Document Processing

  • Extract text content from the user's document using appropriate extraction method based on file type (PDF, DOCX, PPTX)
  • Store the original filename and file format for later use (DOCX files will receive annotated versions)
  • Clean the extracted text by removing chart data, axis labels, table values, and purely quantitative figures
  • Preserve all written content: headlines, paragraphs, footnotes, disclaimers, source attributions, and numbers within sentences
  • Format cleaned text with clear section headers for evaluation

2. Document Classification Using Classification Guide

  • Read references/17_document_classification_guide.md for document type identification guidance
  • Follow the Classification Process outlined in the guide (Steps 1-3)
  • Analyze document format, purpose, target audience, content type, and issuer context
  • Classify document into appropriate category from the guide (A through N)
  • CRITICAL STOP CONDITIONS: If document classification is ambiguous or unclear based on guide's MANDATORY STOP CONDITIONS, STOP and ASK user for clarification before proceeding
  • Use the guide's mapping tables to identify which regulatory frameworks apply to this document type
  • Document your classification decision and reasoning

3. Regulatory Framework Selection

  • Based on document classification from Step 2 and the Classification Guide mappings:
  • Identify REQUIRED frameworks (marked with ✅ in guide)
  • Identify CONDITIONAL frameworks (marked with 🔸 in guide) and verify conditions are met
  • Load the appropriate rule md files.
  • For each selected regulatory framework file, read the "Applicability" section to confirm:
  • Document type matches "Document Types Covered" list
  • Content contains "Key Triggers" mentioned in the section
  • Overall purpose and audience align with framework scope
  • Create final list of applicable vs. non-applicable regulatory frameworks with detailed justification

4. Systematic Framework Evaluation with Entity Scope Filtering

  • TWO-GATE RULE EVALUATION (MANDATORY):

Every rule must pass through two gates before scoring:

GATE 1: Entity Scope — Does this rule apply to this entity type? - Read rule opening sentence and description for entity qualifiers (FINRA member firm, broker-dealer, investment company, registered investment adviser) - Verify entity type matches rule scope (cross-reference Applicability section) - If NO MATCH → Mark "Not Triggered (Entity Type Scope)" and skip to next rule - Examples: MKT.29: "ranking or award given to a FINRA member firm or an investment company" → RIA = NO MATCH FINRA 2210: "communications with the public" from broker-dealer → RIA = NO MATCH SEC 206(4)-1: investment adviser marketing → RIA = MATCH

**GATE 2: Content Triggers** — Does document content activate this rule?
  • Check rule's listed triggers against document content
  • If NO TRIGGERS → Mark "Not Triggered (Content)" and skip to next rule

  • Example: MKT.1 (options yield) triggers only if options strategies discussed

    SCORING: Only proceed to compliance scoring if rule passes BOTH gates.

  • For each APPLICABLE regulatory framework:

    • Create entity scope filter: which rules in this framework apply to this entity type (RIA/BD/Investment Company)
    • For each entity-applicable rule:
    • Apply Gate 1 verification (already done in filter creation)
    • Apply Gate 2: Check content triggers from rule's "Triggers" section
    • If both gates pass: Score 🟢 (Compliant) / 🟡 (Partially Compliant) / 🔴 (Non-Compliant)
    • If Gate 2 fails: Mark as 'Not Triggered (Content)'
    • For entity-inapplicable rules:
    • Mark as 'Not Triggered (Entity Type Scope)' with brief reason
    • Track exact document text relating to each scored rule evaluation

5. Rule-by-Rule Scoring with Mandatory Citations

  • For each triggered rule in applicable frameworks:

  • CRITICAL: ALWAYS cite the specific rule number in format PREFIX.NUMBER

  • Examples: MKT.5, FINRA_2210.12, SEC_REGD_502.3, GIPS.7, SEC_482.8, REG_BI.4

  • Provide 1-3 sentence explanation for each score, always including rule number

  • Apply professional judgment for contextual and subtle compliance issues

  • For violations (🔴 or 🟡), note specific document text and location for later DOCX annotation

    ``` 🔴 MANDATORY PRE-SCORING VERIFICATION AND OUTPUT FORMAT:

    INTERNAL VERIFICATION PROCESS (Complete before scoring):

    Step 0: Category Verification - Is this an ADVERTISEMENT CONTENT requirement (must appear in the ad)? - Or a FIRM-LEVEL OBLIGATION (internal procedures, cannot verify from ad alone)? - Or an OPERATIONAL REQUIREMENT (business processes, cannot verify from ad alone)? - If firm/operational obligation → Mark "Not Triggered (Cannot Verify from Document)" and skip scoring

    Step 1: Quote the Requirement - Copy the exact text from the rule's "Checklist" section that defines compliance

    Step 2: List What's Explicitly Required - Bullet list: What specific items, disclosures, or elements does the rule checklist explicitly require? - Do NOT add items you think should be required - For each item in your list, write the rule citation that requires it. If no citation exists, DELETE that item from your list.

    Step 3: Check for Exemptions and Assumptions - Does the rule contain "not required if..." or "only required when..." language? - Does the rule contain assumption frameworks (e.g., "assume no conflicts exist when...")? - If yes, do any exemptions or assumptions apply to this document?

    Step 4: Match Document to Checklist - For each required item: Is it present? Yes/No - For each exemption: Does it apply? Yes/No

    Step 5: Score Based ONLY on Steps 1-4 - 🟢 = All required items present (after exemptions/assumptions applied) - 🟡 = Most required items present - 🔴 = Required items missing (after confirming no exemptions apply) - Not Triggered = Content doesn't activate the rule

    MANDATORY PRE-SCORE CHECKPOINT: Before assigning 🟡 or 🔴, verify: □ I can quote exact rule text requiring this element □ This is an advertisement content requirement, not a firm-level obligation □ I checked for exemptions and assumption frameworks □ The document match supports this score

    If any checkbox is unchecked → Score defaults to 🟢 Compliant

    REQUIRED OUTPUT FORMAT IN COMPLIANCE REPORTS:

    For 🟢 Compliant Scores: Brief statement with rule citation. - Example: "MKT.7 — Standardized Performance: 🟢 Compliant. Shows 1-, 5-, 10-year periods with proper disclosures."

    For 🟡 Partially Compliant or 🔴 Non-Compliant Scores:

    YOU MUST INCLUDE THIS VERIFICATION BLOCK IN THE DELIVERABLE REPORT (not just internal analysis):

    Pre-Scoring Verification: - Step 1 — Rule Requirement: [Exact quote from rule's checklist] - Step 2 — Required Items: [Only items rule explicitly requires] - Step 3 — Exemptions: [Exemptions checked; state if any apply] - Step 4 — Document Match: [For each item: ✓ Present / ✗ Absent with evidence] - Step 5 — Score Determination: [Logic from Steps 1-4]

    Document Text: [Quote relevant document section]

    ENFORCEMENT: If this verification block is missing or incomplete in the report for any 🟡 or 🔴 score → The score is invalid and must be corrected to 🟢 Compliant.

    CRITICAL RULE: If you cannot cite specific checklist text requiring something, you cannot mark it non-compliant for lacking it.

    STRICT RULE ADHERENCE PROTOCOL: - When a rule contains an explicit checklist (e.g., "Required disclosures must include..."), evaluate ONLY against those listed items - Do not add interpretive requirements not present in the rule text - If a concern exists that isn't covered by the specific rule's checklist, flag it under the appropriate general standard (e.g., SEC_206(4)_1.2 General Prohibitions, MKT.13 General Standards) rather than creating a technical violation where none exists - When the rule provides explicit scoring criteria (🟢/🟡/🔴), use them literally - Before marking any rule non-compliant, verify: (1) What does the rule's checklist explicitly require? (2) Are all those items present? (3) Is my concern actually in the rule's requirements, or am I adding something? ```

6. Comprehensive Compliance Report Generation

  • Create Document Analysis Section:
  • Document type classification (reference Classification Guide category)
  • Target audience identified (retail, institutional, intermediary)
  • Issuer type identified (RIA, BD, fund company, etc.)
  • List of applicable regulatory frameworks with justification
  • List of non-applicable frameworks with justification
  • Summary of why specific regulations apply to this document
  • Create Executive Summary:
  • Overall compliance posture
  • Total applicable frameworks (out of 21)
  • Total rules evaluated (triggered within applicable frameworks)
  • Critical violations count (🔴)
  • Partial compliance issues count (🟡)
  • Key findings and critical violations
  • Generate Detailed Framework Sections:
  • One section per APPLICABLE regulatory framework
  • MANDATORY FORMAT: 'PREFIX.NUMBER — Title: 🟢/🟡/🔴 Explanation'
  • Include framework overview and why it applies
  • Show all rules as triggered or 'Not Triggered'
  • Create Statistics Table:
  • 🟢/🟡/🔴/Not Triggered counts for each applicable framework
  • List excluded frameworks with brief reason
  • Overall compliance percentage
  • Generate Prioritized Recommendations:
  • Critical issues (🔴) first with specific rule citations
  • Partial compliance issues (🟡) second
  • Each recommendation with rule number: 'Violation of MKT.3' or 'FINRA_2212.8 non-compliant'
  • Prioritize by severity and materiality

7. DOCX Document Annotation with Tracked Changes

  • IF the original input document is a DOCX file:
  • Load scripts/word_document_editor.py - the Word document editing utility
  • Create a working copy of the original DOCX file
  • For each compliance issue found (🔴 or 🟡 ratings):
    • Locate the specific text/section in document that triggered the violation
    • Add a comment using word_document_editor.py's search_and_comment() method
    • Comment format: "RULE_ID: [Violation description]. Severity: [Red/Yellow]. Recommendation: [Brief fix guidance]"
    • Comment author: "Compliance Review Agent"
  • For 🔴 (Red) violations where specific text changes would resolve:
    • Use word_document_editor.py's search_replace_track_changes() method
    • Add suggested corrections with track changes enabled
    • Track changes author: "Compliance Review Agent"
  • Preserve all original document content, formatting, and styles
  • Save annotated document as: "[original_name]_COMPLIANCE_REVIEW.docx"
  • Generate summary: Number of comments added, number of tracked changes suggested
  • Note: This step only applies to DOCX input files, not PDF or PPTX

Completion

  • Deliver complete compliance report with:
  • Document Analysis section with classification from guide
  • All applicable rules evaluated or marked as Not Triggered
  • Frameworks excluded with justification
  • Comprehensive statistics showing coverage across 21 frameworks
  • Actionable prioritized recommendations with rule number citations
  • IF DOCX file: Also deliver the annotated DOCX file with:
  • Comments on all compliance issues (🔴 and 🟡)
  • Tracked changes with suggested corrections for violations
  • Summary of annotations made
  • Ensure report is professionally formatted with clear section headers and consistent numbering

Shared as is, with no warranty. Not investment, legal or tax advice. See Disclosures.