Back to skill

Security audit

Addiction Treatment Marketing Kit

Security checks for vulnerabilities and agentic risk

Overview

This prompt-only skill is coherent for healthcare marketing, but one template can produce false certification and credential claims in regulated addiction-treatment content.

Review before installing or using in production. The skill does not appear to run code or access data, but do not rely on its compliance claims as-is; require human compliance/legal review and fix the unconditional certification footer so every credential claim is rendered only when active, verified, and applicable.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
prompts/01-patient-acquisition-campaigns.md:20
Finding

Mandatory Footer Can Generate False Certification and Credential Claims

Content
View full analysis

Vulnerability Details

File Location: prompts/01-patient-acquisition-campaigns.md, lines 20, 42, and 100–102
Vulnerability Type: Inconsistent validation and unconditional rendering of regulated credentials
Risk Level: Medium

Vulnerable Code Snippet

text
**Gate 3 — LegitScript notation:** If any digital delivery channel is mentioned (email with click-to-website, SMS with URL, etc.), flag that paid digital advertising requires active LegitScript certification separately.
text
- LegitScript status: [ACTIVE (cert #: [NUMBER], expiration: [DATE]) / PENDING / NOT CERTIFIED]
text
## MANDATORY COMPLIANCE FOOTER (include on all outputs)

*[FACILITY_NAME] is licensed by the Nevada Division of Public and Behavioral Health (DPBH License #[DPBH_LICENSE]). [ACCREDITATION_BODY] accredited (#[ACCREDITATION_NUMBER]). LegitScript certified (#[LEGITSCRIPT_NUMBER]). Medical Director: [MD_NAME], MD, [BOARD_CERT], NV Medical Board #[NV_MB_NUMBER]. This communication does not constitute a guarantee of treatment outcomes. Individual results vary based on diagnosis, co-occurring conditions, treatment engagement, and post-treatment support.*

Technical Analysis

The prompt explicitly accepts PENDING and NOT CERTIFIED as possible LegitScript input states. Gate 3 only requires a warning when certain digital delivery channels are involved; it does not block offline campaigns for facilities without active certification.

However, the prompt later requires the same footer on every output. That footer unconditionally states that the facility is “LegitScript certified” and also unconditionally asserts accreditation and medical-director credentials. These claims are not conditionally rendered according to verification status.

This creates a validation-to-output inconsistency: an input may truthfully indicate that certification is pending or absent, yet the generated output is instructed to claim active certification. Similar false claims may ...[truncated 1600 chars]

Remediation
View remediation

Remediation Suggestions

  1. Replace the unconditional footer with status-aware conditional rendering:

    • Include “LegitScript certified” only when the status is explicitly ACTIVE.
    • Require a nonempty certification number and unexpired date.
    • Omit the claim or accurately state pending or not certified for other states.
  2. Add equivalent validation gates for every footer credential:

    • Active DPBH license and matching service scope.
    • Current accreditation and covered service lines.
    • Verified medical-director license and board certification.
    • Correct identity and credential-number format.
  3. Fail closed when required verification data is missing, contradictory, expired, or malformed. Do not infer that placeholder values represent verified credentials.

  4. Separate footer templates by verified status. For example:

    • Active certification: include the verified certification claim and number.
    • Pending certification: state that certification is pending only if appropriate.
    • No certification: omit all certification language and warn that the content cannot be used as paid addiction-treatment advertising.
  5. Add explicit prompt tests for:

    • Active certification with valid number and expiration date.
    • Pending certification.
    • Expired certification.
    • No certification.
    • Missing or malformed certification number.
    • Missing or unverified accreditation and medical credentials.
  6. Require a final consistency check that compares every generated credential claim against the supplied, verified input before returning content.

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (6)

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 40)May include surrounding context.

md
- Any paid advertising content claiming to be "certified," "accredited," or "approved" when LegitScript certification is pending or expired
- Organic SEO content that uses paid-ad framing ("immediate intake," "available beds now") that triggers platform review as potential advertising policy evasion

**Why it matters:** Google suspended ALL addiction treatment advertising in September 2017 citing widespread fraud, patient brokering, and consumer harm. After 14 months, Google reinstated the category in October 2018 — but only for LegitScript-certified providers. Meta/Facebook followed with an identical requirement. LegitScript certification costs $995–$1,500 annually and requires: state facility license documentation, proof of accreditation (CARF/JCAHO or equivalent), licensed clinical staff documentation, no patient brokering arrangements, compliance with NAATP Code of Ethics, and a site inspection for residential programs. Every AI marketing tool generates Google and Meta ad copy for treatment centers without asking whether the client is LegitScript-certified — because LegitScript certification is a platform enforcement mechanism that postdates all general AI training data. Every uncertified treatment center that runs this AI-generated ad copy has their account suspended within days of launch.

**Required input:** LegitScript certification status (active/pending/not-certified) + LegitScript account number if active + state facility license number + CARF or Joint Commission accreditation (yes/no, accreditation number) + services covered by certification

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 54)May include surrounding context.

md
- "We coordinate directly with your PCP and specialists" — inter-provider record sharing for SUD patients requires Part 2-specific written consent (more restrictive than standard HIPAA authorization)
- "Alumni network" communications that imply former patients can be identified or contacted based on their treatment history without individually documented consent
- Before/after testimonials from identifiable alumni without a 42 CFR Part 2 compliant written authorization that explicitly permits marketing use (separate from treatment consent)
- "Our staff will contact your employer's EAP" — EAP coordination requires Part 2 consent; employer notification of SUD treatment without consent is a federal violation
- "We accept [specific named employer] EAP" communications that imply the facility shares treatment information with named employers without consent documentation
- Insurance verification communications that share patient identity + treatment service category with payers without documented Part 2 consent
- GBP or social media responses that confirm or imply a specific individual received addiction treatment ("We're glad John's recovery is going well")

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 55)May include surrounding context.

md
- "We coordinate directly with your PCP and specialists" — inter-provider record sharing for SUD patients requires Part 2-specific written consent (more restrictive than standard HIPAA authorization)
- "Alumni network" communications that imply former patients can be identified or contacted based on their treatment history without individually documented consent
- Before/after testimonials from identifiable alumni without a 42 CFR Part 2 compliant written authorization that explicitly permits marketing use (separate from treatment consent)
- "Our staff will contact your employer's EAP" — EAP coordination requires Part 2 consent; employer notification of SUD treatment without consent is a federal violation
- "We accept [specific named employer] EAP" communications that imply the facility shares treatment information with named employers without consent documentation
- Insurance verification communications that share patient identity + treatment service category with payers without documented Part 2 consent
- GBP or social media responses that confirm or imply a specific individual received addiction treatment ("We're glad John's recovery is going well")

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 121)May include surrounding context.

md
- Sober living "guarantees" — NARR-certified sober living homes are peer-support environments, not treatment facilities; marketing them as clinical treatment is misrepresentation
- "Family intervention services" implying coercive intervention without disclosure that the ARISE model and other ethical intervention approaches do not use surprise confrontation tactics condemned by clinical evidence
- "Insurance will cover everything" as a sales tool without benefits verification — balance billing and coverage gap risks must be disclosed
- False insurance network claims: "In-network with [specific plan]" without verification (insurance credentialing changes; in-network status fluctuates)
- "Luxury rehab" claims that imply amenities substitute for clinical programming — FTC has targeted amenity-over-treatment marketing as UDAP-deceptive
- "Teen program" advertising without confirming that adolescent treatment requires additional safeguards: parental consent, CPS/DJJ coordination protocols, DCFS reporting obligations

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
86% confidence
Finding

The template prescribes an alumni outreach review-request sequence in English only and does not instruct the operator to determine the recipient's preferred language or provide equivalent translations. In a behavioral health and addiction-treatment context, this can exclude limited-English-proficiency recipients, undermine informed consent and opt-out comprehension, and create compliance and discrimination risk if outreach is sent to patients who cannot fully understand the communication.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
77% confidence
Finding

This markdown file invites users to call the intake team to schedule a clinical assessment or verify insurance benefits, which implies collection of sensitive health and insurance information. The surrounding skill output provides no explicit warning about sharing personal health information or how such information will be handled, which is a user-facing disclosure gap under the markdown-file warning criterion.

Content

No source excerpt is available for this finding.