Back to skill

Security audit

ORBIT Platform

Security checks for vulnerabilities and agentic risk

Overview

The skill is coherent for building ORBIT, but it gives broad live-platform guidance around credentials, privileged Supabase access, Telegram ingestion, and generated HTML without enough safety scoping.

Install only if you control the ORBIT project and will revise the operational rules before use: do not let agents read or echo credential files, replace blanket service_role use with least-privilege server-side credentials, authenticate and validate Telegram webhooks, add privacy/retention controls for stored user data, and escape or sandbox all generated HTML presentations.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • 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
Findings (3)

T09 · Insecure Skill Coding Practices

Error
Location
references/exemplos-agentes.md:142
Finding

Unauthenticated Telegram Webhook Permits Forged Job Submission

Content
View full analysis
{ const update = req.body as TelegramUpdate const message = update.message if (!message?.text) return reply.send({ ok: true }) // Upsert user_profile const { data: user } = await supabase .from('user_profiles') .upsert({ telegram_id: message.from.id, first_name: message.from.first_name, username: message.from.username, default_chat_id: message.chat.id }, { onConflict: 'telegram_id' }) .select().single() // Criar command const { data: command } = await supabase .from('commands') .insert({ user_id: user.id, payload: { text: message.text, source: 'telegram', chat_id: message.chat.id } }).select().single() // Criar job const { data: job } = await supabase .from('jobs') .insert({ command_id: command.id, chat_id: message.chat.id, status: 'pending' }).select().single() // Enfileirar await supabase.rpc('push_intent_job', { p_job_id: job.id }) // Resposta imediata ao usuário await sendTelegramMessage(message.chat.id, '🔄 Processando sua solicitação...') return reply.send({ ok: true }) }) ``` ### Technical Analysis The endpoint trusts the submitted request body merely by casting it to `TelegramUpdate`. It does not verify Telegram's `X-Telegram-Bot-Api-Secret-Token` header, authenticate the caller through another mechanism, validate the request schema, or reject replayed updates. An arbitrary Internet client able to reach this route can therefore impersonate Telegram users and chats. The handler uses attacker-controlled identifiers to update `user_profiles`, creates a command containing attacker-controlled prompt text, creates a job, and invokes the queue RPC. The ...[truncated 1267 chars]
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Error
Location
SKILL.md:23
Finding

Skill Directs Agents to Access Credential Files and Always Use the Supabase Service-Role Key

Content
View full analysis
Remediation
View remediation

T09 · Insecure Skill Coding Practices

Error
Location
references/exemplos-agentes.md:192
Finding

Unescaped Dossier Content Enables Stored HTML Injection and Cross-Site Scripting

Content
View full analysis
` ${dossier.titulo} /* Estilos inline — sem CDN externo */ :root { --primary: #6366f1; --bg: #0f0f1a; --text: #e2e8f0; } body { background: var(--bg); color: var(--text); font-family: system-ui; margin: 0; } .hero { padding: 4rem 2rem; background: linear-gradient(135deg, #1a1a3e, #0f0f1a); } .hero h1 { font-size: 2.5rem; color: #818cf8; } .section { padding: 2rem; max-width: 900px; margin: 0 auto; } .swot-grid { display: grid; grid-template-columns: 1fr 1fr; gap: 1rem; } .swot-card { padding: 1.5rem; border-radius: 12px; } .forcas { background: #0d2b1a; border: 1px solid #22c55e; } .fraquezas { background: #2b0d0d; border: 1px solid #ef4444; } .oportunidades { background: #0d1f2b; border: 1px solid #3b82f6; } .ameacas { background: #2b1a0d; border: 1px solid #f59e0b; }

${dossier.titulo}

${dossier.resumo_executivo?.substring(0, 200)}...

` ``` ### Technical Analysis The template inserts `dossier.titulo` and `dossier.resumo_executivo` directly into HTML contexts without encoding or sanitization. These fields are generated from a pipeline that begins with untrusted user text and incorporates external research content. Consequently, they cannot safely be treated as trusted markup. An attacker can attempt to make a dossier field contain closing tags, new elements, event handlers, dangerous URLs, or script-bearing markup. For example, a title containing `
Remediation
View remediation
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (11)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The trigger phrases are broad and ambiguously scoped, making it more likely this powerful infrastructure skill is invoked in contexts beyond the user's intent. Because the skill includes guidance touching credentials, database operations, migrations, and agent orchestration, accidental activation can cause over-privileged assistance, unsafe code generation, or inappropriate handling of sensitive systems.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill points directly to .env and a credentials file, effectively steering the agent toward sensitive secret locations without any warning or restriction. In a hostile or confused interaction, this increases the chance of secret discovery, summarization, or accidental disclosure, especially given the surrounding instructions to operate a live platform.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill explicitly instructs operators to always use the Supabase service_role key, which grants broad administrative access and bypasses normal row-level security controls. In an agentic platform that processes external inputs from Telegram, dashboards, and multiple handoffs, this blanket guidance increases the blast radius of prompt abuse, coding mistakes, or data-destructive operations.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The instruction "Idioma padrão: Português do Brasil em toda comunicação com o usuário" imposes a language/locale policy unconditionally. There is no indication that users may select another language or that the locale restriction is optional or justified as region-specific.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The instruction 'Sempre use Português do Brasil' imposes a fixed language policy. The provided text does not offer opt-in, fallback, or justification that this is a region-specific tool, so it matches the language/locale policy violation criteria.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The webhook example persists Telegram identifiers, chat IDs, usernames, and raw message text to Supabase without any visible consent flow, data-minimization controls, or retention/privacy guidance. In the ORBIT context, these fields can contain personal data and potentially sensitive user prompts, so normalizing this pattern in example code increases the risk of privacy violations, over-collection, and accidental downstream exposure.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
50% confidence
Finding

Data is being sent to an external URL. This could be legitimate telemetry or data exfiltration. Manual review is recommended.

Content

Scanner excerpt · references/perfis-agentes.md (reported line 31)May include surrounding context.

md
- **Missão:** Buscar papers, DOI, referencias robustas
- **Input:** Tema ou pergunta
- **Output:** scholarly_research_result (JSON)
- **Ferramentas:** web_search_preview + OpenAlex API (https://api.openalex.org/works?search=)
- **Quality Gate:** ≥ 3 referências com DOI verificável
- **Campos no research_sources:** doi, citation_count, publication_year, authors, institution

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The schema documentation specifies language (pt-BR) for the presentations table, which indicates a fixed locale. Under the policy, forcing a specific language without user opt-in is a natural-language policy violation unless the locale constraint is explicitly justified, which is not present here.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
92% confidence
Finding

The manifest sets the skill language to "pt-BR", and the visible metadata does not indicate that users can choose another language or opt into this locale. Under the policy, forcing a specific language without user choice is a natural-language policy concern unless the locale restriction is explicitly justified.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

This markdown file contains natural-language schema descriptions exclusively in Portuguese, such as field explanations and section titles, but does not state that Portuguese is optional or that the skill is region-specific. Under the language/locale policy, forcing a single language without user opt-in can be a policy concern.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

This markdown file contains natural-language instructions and profile descriptions exclusively in Portuguese. Under the language/locale policy rule, forcing a specific language without user opt-in can be a policy concern when no alternative language option or justification is provided.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.