Back to skill

Security audit

Finanças Donna

Security checks for vulnerabilities and agentic risk

Overview

This appears to be a real personal-finance skill, but it exposes real personal financial details and uses broad triggers that could activate around sensitive files too easily.

Review before installing. This skill is only appropriate for the named person's private finance workspace; remove real personal names, relationships, and amounts before sharing or reuse, and narrow activation to explicit finance commands so it does not read or update sensitive records accidentally.

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
SKILL.md:211
Finding

Plaintext Exposure of Personal and Financial Information

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 211–224
Vulnerability Type: Plaintext sensitive data embedded in Skill configuration
Risk Level: Medium

Vulnerable Code

markdown
### Contexto Pessoal do Evandro
- **Empresa própria / Autônomo** → receitas vêm direto dos clientes (ex: Avanto)
- **NÃO registrar "Retirada Empresa"** → os pagamentos dos clientes JÁ SÃO a receita
- Clientes podem ter % de repasse (ex: 35%) → registrar apenas a parte do Evandro
- **Namorada:** Ana Vitoria Nogueira Vian → PIX enviados = gastos reais, PIX recebidos = contribuições/reembolsos
- **Sogro:** Alexandre Placido Vianna → contribui com parte do aluguel
- **Aluguel na planilha:** registrar apenas o valor LÍQUIDO (aluguel total - contribuição Alexandre). NÃO registrar Alexandre como receita separada
- **Mãe:** Vanessa Cristina da Costa → PIX recebido = reembolso (comprou pra ela), NÃO é receita. PIX enviado = gasto real.
- **Irmã:** Evelin Thomazini Castilho → mesmo padrão da mãe
- **PIX de/para EVANDRO THOMAZINI** = transferência entre contas próprias da empresa. Só contabilizar o RECEBIDO como "Salário / Renda Principal"
- **Pagamentos de cartão de crédito no extrato** = NÃO registrar (já estão nas faturas)
- **Resgates CDB/RDB** = NÃO registrar (movimentação de investimentos)
- **Aluguel:** R$ 2.202,16/mês - Invest Imoveis (categoria Moradia)
- **IPVA:** R$ 276,71/mês (categoria Outras Fixas)

Technical Analysis

The Skill definition embeds personally identifiable and sensitive financial information directly in a distributable plaintext file. The exposed data includes full names and family relationships, recurring financial obligations, a property counterparty, account-transfer behavior, reimbursement rules, and business-income classification details.

Skill instructions are loaded into the Agent's context when the Skill is activated. Consequently, every party with access to the package, reposit ...[truncated 2020 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove all real names, relationships, counterparties, account-owner identifiers, and exact recurring amounts from SKILL.md.
  2. Replace personal examples with synthetic placeholders such as PARTNER_NAME, RELATIVE_NAME, and MONTHLY_RENT_AMOUNT.
  3. Store user-specific classification rules in a separate configuration file that is excluded from source control and Skill distribution.
  4. Protect that configuration using strict filesystem permissions and encryption at rest where supported.
  5. Load personal configuration only when required and only with explicit user consent.
  6. Minimize the data supplied to the Agent context; provide only the rule necessary for the current transaction.
  7. Add secret and sensitive-data scanning to the repository and release pipeline.
  8. Review repository history and distributed artifacts for prior exposure. If the file was published, remove sensitive history where feasible and notify affected individuals as appropriate.
  9. Add tests ensuring that the generic Skill operates without relying on identities or fixed financial amounts.
Vulnerability Patterns
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (2)

Vague Triggers

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The auto-activation triggers are extremely broad and include common everyday words like 'comprei', 'paguei', 'conta', 'resumo' and month references, which can cause the finance skill to activate in unrelated conversations. Because the skill instructs the agent to read and update financial files once activated, accidental invocation can expose sensitive financial data or cause unintended modifications to records.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The file-processing triggers 'fatura', 'extrato', 'cartão', and 'banco' are too generic and can match ordinary discussion or unrelated documents. In this skill, those triggers can lead the agent to read and interpret uploaded files, increasing the risk of processing the wrong document or exposing sensitive banking data without sufficiently specific user intent.

Content

No source excerpt is available for this finding.