Compliance-Aware Content Review

The SEC Marketing Rule applies to registered investment advisers. The skill screens content against the seven general prohibitions:

1. Untrue statements of material fact -- Flag any factual claim that is not substantiated by source data in the workspace.
2. Unsubstantiated material claims -- Require that every material statement of fact has a traceable data source or citation.
3. Untrue or misleading implications -- Detect language that could create a false impression even if technically accurate in isolation.
4. Failure to disclose material facts -- Check that content mentioning performance or investment characteristics includes all material context.
5. Misleading use of cherry-picked performance -- Detect selective time-period presentation; require all standard reporting periods.
6. Statements materially misleading to the intended audience -- Consider whether claims appropriate for institutional investors appear in retail-facing content.
7. Otherwise materially misleading statements -- Catch-all for content that creates unjustified expectations.

Performance Presentation Standards

• Gross and net performance must receive equal prominence when both are shown.
• Performance must include 1-year, 5-year, and 10-year (or since-inception) periods.
• Benchmarks must accompany any performance comparison.
• Hypothetical and back-tested performance requires specific disclosures.
• Extracted performance must include total portfolio results.

Testimonial and Endorsement Requirements

• Cash and non-cash compensation must be disclosed.
• Material conflicts of interest must be stated.
• Content from current clients must note that client status exists.
• Testimonials must not be misleading when read in context.

FINRA Rule 2210 Compliance

FINRA Rule 2210 governs communications with the public by broker-dealers.

Content Classification

The skill classifies each piece of content into one of three FINRA categories:

Fair and Balanced Standard

• Every statement of benefit must be balanced with associated risks.
• Claims must not exaggerate potential returns or minimize risks.
• Projections and forecasts are prohibited for retail communications.
• Comparisons to other products must be fair and substantiated.
• Past performance language must include "past performance does not guarantee future results" or equivalent approved language.

Filing Requirements

Flag content requiring FINRA pre-filing:

• Communications from firms in their first year of membership.
• Options-related content (pre-use filing with OCC).
• CMO/structured product communications.
• Leveraged and inverse ETF materials.
• Communications about registered investment companies (mutual funds, ETFs).

FCA Financial Promotions Compliance

The FCA regulates financial promotions in the United Kingdom under FSMA 2000 s.21.

Clear, Fair, and Not Misleading Standard

All financial promotions must be:

• Clear -- Key information is prominent and understandable to the target audience.
• Fair -- Benefits and risks receive balanced treatment; no cherry-picking.
• Not misleading -- Nothing that could create a false or deceptive impression.

Risk Warning Requirements

• Specific risk warnings for each product category (UCITS, AIFs, structured deposits).
• Capital-at-risk warning when applicable.
• Past performance warning: "Past performance is not a reliable indicator of future results."
• Complexity warnings for products classified as complex under MiFID II.

Target Market Assessment

• Content must be appropriate for the identified target market.
• Retail investor materials have stricter clarity requirements than professional.
• High-risk investment warnings required where applicable (as of FCA PS22/10).

Disclosure Management

Disclosure Insertion Logic

Based on content classification, the skill determines which disclosures are required and suggests insertion points:

Disclosure Validation

• Verify all required disclosures are present for the content type.
• Check disclosure placement: must be prominent, not buried in footnotes for key items.
• Validate disclosure language matches firm-approved templates.
• Confirm disclosures are current (date-sensitive disclaimers updated).

Customization

Disclosure templates support per-firm customization. Each organization maintains its own approved disclosure language in references/disclosure_templates.md. The skill uses these templates as the basis for insertion and validation, falling back to regulatory-standard language if firm-specific templates are not configured.

Content Classification

The skill classifies incoming content along multiple dimensions:

1. Regulatory jurisdiction -- SEC (US-registered advisers), FINRA (US broker-dealers), FCA (UK), or multiple.
2. Audience type -- Institutional vs. retail vs. correspondence.
3. Content type -- Performance, testimonial, educational, promotional, commentary.
4. Risk level -- High (definite violation), Medium (likely issue), Low (potential concern), Info (best-practice suggestion).
5. Filing requirement -- Pre-use filing, post-use filing, or supervision only.

Archival Tagging

All reviewed content is tagged with archival metadata for regulatory retention:

• Content ID -- Unique identifier for the reviewed piece.
• Review timestamp -- ISO 8601 datetime of the review.
• Reviewer -- "automated-first-pass" (always; human reviewer added later).
• Retention period -- Per SEC Rule 17a-4 and FINRA requirements (typically 3-6 years depending on content type).
• WORM flag -- Whether the content must be stored in Write-Once-Read-Many format.
• Filing status -- Whether FINRA filing is required and current status.
• Classification -- The content classification determined during review.

See references/archival_requirements.md for retention schedules.

Input / Output Data Contracts

Inputs

Outputs

Output Schema: review_report.json

{
  "review_id": "string (UUID)",
  "review_timestamp": "string (ISO 8601)",
  "content_source": "string (file path)",
  "overall_status": "PASS | FAIL | WARNING",
  "advisory_notice": "This is an advisory first-pass review, not compliance certification.",
  "issues": [
    {
      "issue_id": "string",
      "severity": "HIGH | MEDIUM | LOW | INFO",
      "category": "SEC | FINRA | FCA | DISCLOSURE | ARCHIVAL",
      "rule_citation": "string (e.g., 'SEC Rule 206(4)-1(a)(2)')",
      "description": "string",
      "location": "string (content location reference)",
      "remediation": "string (suggested fix)",
      "requires_human_review": true
    }
  ],
  "disclosures_required": ["string"],
  "disclosures_present": ["string"],
  "disclosures_missing": ["string"],
  "classification": {
    "jurisdiction": ["SEC", "FINRA", "FCA"],
    "audience": "INSTITUTIONAL | RETAIL | CORRESPONDENCE",
    "content_type": "string",
    "filing_required": "PRE_USE | POST_USE | NONE"
  },
  "archival_metadata": {
    "retention_period_years": 0,
    "worm_required": true,
    "content_hash": "string (SHA-256)"
  }
}

Cross-Skill Integration

Compliance review is the mandatory terminal gate in financial services workflows. Every skill that produces customer-facing content must route through compliance-review before distribution when the workspace is tagged as financial services:

• reporting -- Executive summaries with performance claims.
• email-analytics -- Email content referencing investment returns or products.
• paid-media -- Ad copy for financial products or services.
• seo-content -- Website content making investment-related claims.
• social-analytics -- Social media posts about financial performance.
• attribution-analysis -- Reports containing performance data that may be used in marketing.

The skill reads content from other skills' output directories and writes compliance review results to workspace/compliance/. It does not block other skills from running; it operates as a post-production gate.

Development Guidelines

1. Advisory only -- Compliance review must function as an advisory tool, not an automated approval system. Always recommend human compliance officer final review. Every output must state: "This is an advisory first-pass review, not compliance certification."

2. Versioned rule database -- The rule database in references/ must be versioned and updatable without modifying core skill scripts. Reference files carry their own effective dates and version numbers.

3. Precision over recall -- False positive rate must be below 30% to maintain reviewer trust. Prioritize precision over recall for low-severity issues. High- severity violations (superlatives, guarantees) should have near-100% recall.

4. Firm-customizable disclosures -- Disclosure templates must support customization per firm. Each organization has specific approved language that overrides defaults.

5. Severity distinction -- Clearly distinguish between definite violations (superlative claims, guarantees) and potential issues requiring judgment (tone, emphasis). Use HIGH severity only for clear rule violations.

6. Archival compatibility -- Archival tagging must produce metadata compatible with common compliance archival systems (Smarsh, Global Relay, Bloomberg Vault).

7. No authoritative claims -- Never claim compliance decisions are authoritative. Always label output as "first-pass review" requiring human confirmation.

8. Audit trail -- The review log must be append-only and immutable once written. Each entry includes timestamp, content hash, findings, and reviewer identity.

9. Modular scripts -- Scripts handle deterministic computation (pattern matching, validation, insertion, tagging). The LLM handles nuanced interpretation and contextual judgment. Keep these responsibilities clearly separated.

10. Fail-safe defaults -- When in doubt, flag for human review rather than passing content silently. It is better to over-flag than to miss a genuine violation.

Scripts

Reference Files