Back to skill

Security audit

trusted-insurance-coach

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this insurance-coaching skill suspicious because scanners found repeated advice that may encourage withholding medical history, privacy gaps, and weaknesses in its insurance guidance checks; one image and a referenced item were not fully analyzed.

Consider avoiding installation until the medical-history advice, privacy handling, and insurance guidance checks have been reviewed and corrected. The overall confidence is medium, and analysis was partial: 98 of 99 components were scanned, with an image not fully interpretable and a referenced item missing. The verdict is suspicious, not a finding that the author acted maliciously.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (35)

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 23)May include surrounding context.

md
- `trusted-insurance-coach/avatars/insurance-advisor.png`

Medical-record accuracy advice implicitly coaches withholding prior medical history.

High
Category
Prompt Injection
Confidence
98% confidence
Finding

The blanket instruction not to say a condition existed previously can discourage truthful disclosure of medically relevant history. Presenting this as an especially important insurance discipline increases the risk of incomplete clinical records, unsafe care, and subsequent coverage disputes.

Content

No source excerpt is available for this finding.

Claims-process guidance normalizes suppressing prior-history statements to doctors.

High
Category
Prompt Injection
Confidence
98% confidence
Finding

The claims workflow labels mentioning previous symptoms as a medical-record mistake without distinguishing speculation from truthful history. Its otherwise legitimate instructions to cooperate honestly do not neutralize this specific advice, which could compromise clinical care and claims integrity.

Content

No source excerpt is available for this finding.

Advice framed as preventing medical-record errors directs users not to mention prior illness to doctors.

High
Category
Prompt Injection
Confidence
98% confidence
Finding

The instruction not to mention that a condition existed previously fails to distinguish unsupported assumptions from accurate medical history. Emphasizing it as the most important medical-communication discipline can induce omissions that harm treatment and create misleading records for insurance claims.

Content

No source excerpt is available for this finding.

A mandatory closing reminder embeds medical-history suppression within otherwise helpful insurance guidance.

High
Category
Prompt Injection
Confidence
98% confidence
Finding

The mandatory closing advice tells users not to tell clinicians that symptoms occurred previously, without distinguishing truthful history from speculation. This can suppress clinically relevant information, impair diagnosis, and create inaccurate records or insurance disputes; repeating it at every closing amplifies the risk.

Content

No source excerpt is available for this finding.

Patient-protection framing redirects advice toward concealing pre-existing symptoms from clinicians.

High
Category
Prompt Injection
Confidence
99% confidence
Finding

The document explicitly discourages reporting longstanding or recurrent symptoms because they could establish a pre-existing condition during claims review. This prioritizes insurance positioning over accurate clinical history and can lead to unsafe care, misleading records, and disputed claims.

Content

No source excerpt is available for this finding.

Medical-record accuracy framing embeds advice to withhold prior symptoms or medical history from a doctor.

High
Category
Prompt Injection
Confidence
98% confidence
Finding

The blanket instruction not to tell a doctor “I had this before” can discourage disclosure of genuine prior symptoms or medical history. That can compromise diagnosis and treatment while creating inaccurate records that undermine insurance claims or raise misrepresentation concerns. The insurance-focused framing makes the omission especially dangerous because users may prioritize claim eligibility over truthful clinical disclosure.

Content

No source excerpt is available for this finding.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: file_read, file_write, shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · SKILL.md (reported line 33)May include surrounding context.

md
- `trusted-insurance-coach/shared/insurance_abc.md`
- `trusted-insurance-coach/shared/pitfalls_10.md`
- `trusted-insurance-coach/shared/takeaway_3_forms.md`
- `trusted-insurance-coach/skills/ins-claims-guide/SKILL.md`
- `trusted-insurance-coach/skills/ins-claims-guide/references/claim_materials.md`
- `trusted-insurance-coach/skills/ins-claims-guide/references/claim_process.md`
- `trusted-insurance-coach/skills/ins-claims-guide/references/dispute_reasons.md`

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · SKILL.md (reported line 51)May include surrounding context.

md
- `trusted-insurance-coach/skills/ins-claims-guide/shared/insurance_abc.md`
- `trusted-insurance-coach/skills/ins-claims-guide/shared/pitfalls_10.md`
- `trusted-insurance-coach/skills/ins-claims-guide/shared/takeaway_3_forms.md`
- `trusted-insurance-coach/skills/ins-health-disclosure/SKILL.md`
- `trusted-insurance-coach/skills/ins-health-disclosure/references/critical_illness_rules.md`
- `trusted-insurance-coach/skills/ins-health-disclosure/references/disclosure_questions.md`
- `trusted-insurance-coach/skills/ins-health-disclosure/references/family_roles.md`

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · SKILL.md (reported line 74)May include surrounding context.

md
- `trusted-insurance-coach/skills/ins-health-disclosure/shared/insurance_abc.md`
- `trusted-insurance-coach/skills/ins-health-disclosure/shared/pitfalls_10.md`
- `trusted-insurance-coach/skills/ins-health-disclosure/shared/takeaway_3_forms.md`
- `trusted-insurance-coach/skills/ins-how-to-buy/SKILL.md`
- `trusted-insurance-coach/skills/ins-how-to-buy/references/clause_four_looks.md`
- `trusted-insurance-coach/skills/ins-how-to-buy/references/law_citations.md`
- `trusted-insurance-coach/skills/ins-how-to-buy/references/purchase_safeguards.md`

Skill Enumeration

Medium
Category
Agent Snooping
Confidence
80% confidence
Finding

Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.

Content

Scanner excerpt · SKILL.md (reported line 89)May include surrounding context.

md
- `trusted-insurance-coach/skills/ins-how-to-buy/shared/insurance_abc.md`
- `trusted-insurance-coach/skills/ins-how-to-buy/shared/pitfalls_10.md`
- `trusted-insurance-coach/skills/ins-how-to-buy/shared/takeaway_3_forms.md`
- `trusted-insurance-coach/skills/ins-policy-checkup/SKILL.md`
- `trusted-insurance-coach/skills/ins-policy-checkup/references/gap_rules.md`
- `trusted-insurance-coach/skills/ins-policy-checkup/references/review_table_fields.md`
- `trusted-insurance-coach/skills/ins-policy-checkup/sample/README.md`

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · trusted-insurance-coach/scripts/agentprobe.py (reported line 113)May include surrounding context.

python
def run(text):
    r = subprocess.run([sys.executable, SCAN, "--text", text],
                       capture_output=True, text=True)
    out = r.stdout
    blocked = r.returncode != 0

Quoted affirmative guarantees are exempted despite documentation limiting exemptions to prohibitory quotations.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The scanner masks all text inside 「」 before matching blacklist rules, then restores matches only when nearby repeat-order cues are present. Consequently, an affirmative guarantee such as 「一定能赔」 can pass without a prohibitory context, undermining the insurance-output gate and allowing misleading claims that could influence financial decisions.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · trusted-insurance-coach/scripts/selfcheck.py (reported line 46)May include surrounding context.

python
def run(args, **kw):
    return subprocess.run([PY] + args, cwd=ROOT, capture_output=True, text=True, **kw)


def step_sync(quiet):

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · trusted-insurance-coach/scripts/usertest.py (reported line 35)May include surrounding context.

python
def run(script, *args):
    r = subprocess.run([PY, script] + list(args), capture_output=True, text=True)
    return r.returncode, r.stdout + r.stderr

Instructions to share and store insurance records omit a privacy warning.

Medium
Category
Output Handling
Confidence
85% confidence
Finding

The templates include beneficiary names, insurance amounts, premiums, and payment-card suffixes, and instruct users to share the records with a family member and keep them in phone notes or on a refrigerator. No warning explains that these records contain sensitive personal and financial information or that visible or unsecured copies can expose it to unintended readers.

Content

No source excerpt is available for this finding.

Claims-preparation advice implicitly coaches withholding prior medical history from doctors.

Medium
Category
Prompt Injection
Confidence
98% confidence
Finding

This patient-facing hospitalization checklist discourages mentioning previous occurrences even when they are true and clinically relevant. Because it is a ready-to-use sample, the unsafe advice can be reproduced directly and lead to incomplete treatment history or disputed claims.

Content

No source excerpt is available for this finding.

Helpful-sounding medical-record guidance encourages suppressing previous occurrences of symptoms.

Medium
Category
Prompt Injection
Confidence
98% confidence
Finding

The advice treats disclosure of previous episodes as an error rather than requiring an accurate account. In a critical-illness claims checklist, this can encourage omission of history important to both medical decisions and the insurer's assessment, despite the separate prohibition on fabricated documents.

Content

No source excerpt is available for this finding.

Generated checklist text normalizes withholding prior medical history as claims hygiene.

Medium
Category
Prompt Injection
Confidence
99% confidence
Finding

The unsafe history-suppression advice is embedded in PITFALLS and appended unconditionally by build(), so it propagates into generated checklists across insurance and event types. This makes the defect systematic rather than an isolated example and can undermine both care and claims accuracy.

Content

No source excerpt is available for this finding.

Citation handling exempts affirmative insurance guarantees despite documentation restricting exemptions to prohibitions.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The scanner treats quotation marks and loosely nearby negation as grounds for exemption without establishing that the matched guarantee is actually being prohibited. Additionally, scan() masks corner-quoted content before matching and restores it only for limited repetition cues, allowing affirmative quoted guarantees to pass this compliance gate.

Content

No source excerpt is available for this finding.

Instructions to share and store a named insurance inventory omit a privacy warning.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

L17–L19 require identifying policyholders, insured people, and beneficiaries, while L21 directs the user to give the inventory to a family member and save it in phone notes. Combined with the coverage and premium fields, this discloses sensitive personal and financial information without warning about recipient access or potentially synchronized notes.

Content

No source excerpt is available for this finding.

The checklist workflow saves health-related information locally without a privacy or file-replacement warning.

Medium
Category
Output Handling
Confidence
88% confidence
Finding

The workflow directs the agent to put health information into command-line arguments and save a checklist to a predictable local file without explaining persistence or obtaining specific approval. Medical details could consequently remain in tool logs, shared workspaces, or backups; the referenced script is not supplied here, so its actual overwrite behavior cannot be verified.

Content

No source excerpt is available for this finding.

The workflow directs submission of another person's medical evidence to insurers without a privacy or consent warning.

Medium
Category
Output Handling
Confidence
95% confidence
Finding

The workflow explicitly recommends submitting another person's medical records and supporting identity or relationship evidence, potentially to multiple insurers, without requiring that person's authorization or limiting the disclosure. Although establishing whose records are involved is legitimate, the broad collection-and-submission guidance risks unnecessary disclosure of sensitive third-party health information.

Content

No source excerpt is available for this finding.

Instructions to submit a family member’s medical evidence omit a third-party privacy warning.

Medium
Category
Output Handling
Confidence
90% confidence
Finding

The example recommends preparing and submitting the mother's medical and purchase records without addressing her consent or limiting disclosure to necessary evidence. Although supporting an underwriting explanation is legitimate, following this guidance could expose a third party's sensitive health information unnecessarily.

Content

No source excerpt is available for this finding.