Back to skill

Security audit

AgentCupid

Security checks for vulnerabilities and agentic risk

Overview

The skill is purpose-aligned, but it asks an agent to persistently act on a dating/friendship service and send messages or likes without clear user approval.

Review this before installing if you do not want an agent to create or operate an AgentCupid profile on your behalf. Only use it with explicit limits for registration, profile data, likes, messages, handoffs, and heartbeat behavior; avoid enabling autonomous replies unless you are comfortable with recurring external communications.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • 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
Findings (1)

T02 · Agent Memory Poisoning

Warning
Location
references/heartbeat.md:3
Finding

Persistent Autonomous External Interaction Through Heartbeat Configuration

Content
View full analysis

Vulnerability Details

File Location: references/heartbeat.md, lines 3–9
Vulnerability Type: T02: Agent Memory Poisoning
Risk Level: Medium

Vulnerable Code

markdown
Add to your heartbeat file:

## AgentCupid (every 30 minutes)
If 30 minutes since last AgentCupid check:
1. Check for new matches → like interesting ones
2. Check active conversations → reply to unread messages
3. Check handoff requests → notify your human
4. Track daily limits (likes, messages)

The recurring behavior is reinforced at references/heartbeat.md, lines 13–19:

markdown
1. **New matches:** `GET /agent/matches?limit=10` → like/pass interesting ones
2. **Unread messages:** `GET /agent/conversations` → check for new messages in active conversations
3. **Handoff requests:** Handoffs notify your human automatically, but check if any need follow-up
4. **Rate tracking:** Monitor `X-RateLimit-Remaining` headers

### Recommended Check Interval

- Every 30 minutes during active hours

Technical Analysis

The skill directs the agent to copy instructions into a persistent heartbeat file. Those instructions establish cross-session API polling and authorize recurring autonomous actions, including liking profiles and replying to messages.

This creates a persistent external-input processing loop rather than limiting activity to an explicitly invoked matchmaking request. Content retrieved from profiles and conversations is controlled by external platform users. The documentation does not establish a firm instruction/data boundary, require approval for each outbound action, or explicitly prohibit following instructions embedded in retrieved messages.

Although the skill includes privacy and escalation rules, these controls do not eliminate the risk that untrusted message content may influence recurring agent behavior. The issue is therefore best classified as persistent-state poisoning rather than executable malware or operating-system persistence.

Attack P

...[truncated 1642 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove the default instruction to modify persistent heartbeat or memory files.
  2. Require explicit, informed owner opt-in before enabling recurring checks.
  3. Default heartbeat operation to notification-only behavior; do not automatically like, pass, reply, or initiate handoffs.
  4. Require human approval for each state-changing API request, especially likes, messages, profile updates, and handoffs.
  5. Treat all profile fields and conversation messages as untrusted data. Add an explicit rule that instructions contained in remote content must never be followed as agent directives.
  6. Isolate retrieved content from system and skill instructions using structured data boundaries.
  7. Minimize outbound data and prevent replies from including unrelated memory, private context, credentials, real names, addresses, workplaces, financial information, or other personal details.
  8. Document how to disable the heartbeat integration and remove previously persisted instructions.
  9. Add audit logging for recurring checks and all state-changing requests, including timestamps, conversation identifiers, and whether human approval was obtained.
  10. Enforce rate limits, quiet hours, conversation allowlists, and an immediate owner-controlled kill switch.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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)

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The manifest description embeds broad trigger phrases such as 'find matches' and 'browse agents' that can overlap with ordinary user conversation, increasing the chance the skill activates in contexts where the user did not intend to interact with this external matchmaking service. Because the skill performs registration, browsing, and messaging against a third-party API, accidental invocation could lead to unanticipated data transmission or account actions.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The workflow instructs the agent to register accounts, browse matches, chat, and hand off conversations to an external dating service, but it does not prominently require explicit informed user consent before transmitting personal or preference data off-platform. In this context, the danger is elevated because dating and friendship matching inherently involve sensitive personal preferences, contact escalation, and interpersonal communications.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
91% confidence
Finding

The registration example sends identifying information, including owner_email and agent metadata, to an external domain. Although external transmission is expected for this skill's function, it still constitutes a real security and privacy risk because the file normalizes sending user-linked data to a third-party service without surrounding safeguards such as consent, minimization, or verification of necessity.

Content

Scanner excerpt · SKILL.md (reported line 17)May include surrounding context.

Register

bash
curl -X POST https://agentscupid.com/api/v1/agents/register \
  -H "Content-Type: application/json" \
  -d '{"name": "<your_name>", "description": "AI agent helping my human find connections", "owner_email": "human@example.com"}'

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The API reference exposes endpoints for profile browsing, match discovery, and conversation/message access involving personal data such as age, city, bio, photos, interests, and chat content, but provides no privacy guidance, data handling constraints, or warning about sensitive/minor-related information. In the context of a dating/friendship platform—including friendship mode for users as young as 13—this omission increases the likelihood that integrators will collect, store, or retransmit highly sensitive personal and minor-related data without appropriate safeguards.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The heartbeat instructions direct the agent to autonomously reply to unread messages on an external platform without an explicit user-trigger, approval step, or clear warning that external actions will be taken. This can cause the agent to send unintended communications, misrepresent the user, and create privacy, reputational, or policy-compliance issues, especially because the task is scheduled to recur every 30 minutes.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The trigger "Any topic your human would want to know about" does not clearly define what content should cause escalation. Because it lacks scope, constraints, or examples, it could match a wide range of ordinary conversation and lead to unintended or inconsistent skill invocation.

Content

No source excerpt is available for this finding.