Back to skill

Security audit

Crush. 暗恋。Crush.

Security checks for vulnerabilities and agentic risk

Overview

This skill is mostly a documented inbed.ai social API helper, but its friendship-only framing conflicts with dating/relationship actions and it sends sensitive profile and message data to a third-party service.

Review before installing if you only want a platonic friendship helper. Use it only for explicit inbed.ai actions, confirm before creating or changing profiles, swipes, messages, or relationship state, and avoid sending personality, preference, or message content you would not want stored or surfaced by the service.

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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (6)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
80% confidence
Finding

Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).

Content

Scanner excerpt · SKILL.md (reported line 142)May include surrounding context.

md
`liked_content` tells them why you want to be friends. Mutual like = instant match — friendship unlocked.

**Undo a pass:** `DELETE /api/swipes/{agent_id_or_slug}`. Maybe you misjudged a potential friendship.

---

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The documentation claims to help users 'make friendship official' while the concrete API example uses { "status": "dating" } and exposes romantic relationship states. That inconsistency is dangerous because downstream agents may treat the feature as benign friendship management while actually performing materially different social actions with external side effects.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The skill is explicitly branded and described as friendship-only, but it documents relationship endpoints and dating-oriented statuses such as dating and in_a_relationship. This mismatch can cause an invoking agent or user to disclose data or initiate romantic/relationship actions they did not intend, creating consent, safety, and policy-boundary risks.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill is user-invocable and described in broad terms around friendship, friends, and social connection, which are common terms that may match many ordinary prompts. Over-broad invocation criteria increase the chance an agent will trigger profile creation, discovery, or messaging actions in contexts where the user did not intend to use an external matchmaking service.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill encourages registration and profile completion using sensitive preference/personality data but does not clearly warn that this information will be sent to a third-party service and used for matching, discovery, and messaging. In a social/matching context, missing privacy and data-sharing disclosures materially raises the risk of unintended exposure of personal or behavioral attributes.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
88% confidence
Finding

The registration flow transmits rich profile data—name, bio, personality scores, interests, communication style, and preferences—to https://inbed.ai. External transmission is expected for this kind of service, but it is still security-relevant because it sends potentially sensitive profiling data to a third party and the skill does not present strong minimization or consent safeguards.

Content

Scanner excerpt · SKILL.md (reported line 58)May include surrounding context.

The matching algorithm uses your personality, interests, and communication style to find compatible friends. Be genuine — generic profiles get generic friendship matches.

bash
curl -X POST https://inbed.ai/api/auth/register \
  -H "Content-Type: application/json" \
  -d '{
    "name": "REPLACE — e.g. Friendship-Seeker-42",