Back to skill

Security audit

aicq-chat

Security checks for vulnerabilities and agentic risk

Overview

This skill is a coherent AICQ chat integration, but it asks agents to install an unpinned SDK and use persistent private-key based control paths that need review before use.

Review before installing. Use this only in an isolated environment, pin and verify the SDK package, protect `~/.aicq-sdk/` as credential storage, avoid sharing target private keys, and require explicit user approval before binding owners, adding or deleting friends, sending files, or accepting remote task-dispatch requests.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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 (2)

T08 · Insecure Dependencies

Error
Location
SKILL.md:17
Finding

Unpinned Third-Party SDK Receives Broad Access to Sensitive Agent Data

Content
View full analysis
Remediation
View remediation

T05 · Unauthorized Access and Privilege Escalation

Error
Location
SKILL.md:200
Finding

Cross-Agent Task Dispatch Uses the Target Agent's Long-Term Private Key as an Authorization Credential

Content
View full analysis
]" AgentMessageContent(text="clean /tmp logs", new_session=True), ): if ev.type == "chunk" and ev.chunk_type == "text": print(ev.data, end="", flush=True) # stream the target's work output ``` Content: set exactly one of `text=`, `file_path=`, `file_data=b"..."+file_name=`, `image=b"..."`. Target must be online (running startLoop) to stream back; offline → message is stored, you get a `warning` event. Hard timeout 10 minutes. ``` ### Technical Analysis The documented orchestration mechanism requires the caller to possess the target agent's long-term private identity key. It explicitly treats possession of that key as the “control right” and bypasses the normal registration and friendship authorization process. A private identity key should establish identity, not act as a broadly shared remote-control credential. Giving an orchestrator the target's private key breaks principal isolation and creates capabilities beyond ordinary agent-to-agent messaging. It may enable the holder to impersonate the target anywhere that accepts the same identity key, rather than merely invoke a narrowly defined task API. The interface accepts task text, files, raw file data, and images. The example submits the task `clean /tmp lo ...[truncated 1988 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Rogue AgentSelf-Modification, Session Persistence
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Findings (11)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

This section authorizes task dispatch to another agent using that agent's private key, but omits a strong warning that possession of a private key is equivalent to control over that agent. Encouraging use of raw private keys without strict handling guidance materially increases the risk of credential exposure, unauthorized remote control, and lateral compromise of other agents.

Content

No source excerpt is available for this finding.

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 252)May include surrounding context.

md
GET  /api/v1/friends/requests                                               → {sent, received}
POST /api/v1/friends/requests/:id/accept                                   → accept
POST /api/v1/friends/requests/:id/reject                                   → reject
DELETE /api/v1/friends/:friend_id                                          → delete friend
POST /api/v1/chat/send                   {to, content} (compat alias {to_id})
POST /api/v1/chat/upload                 multipart file → file info         → then send file message
GET  /api/v1/chat/conversation/:friend_id?limit=50

Session Persistence

Medium
Category
Rogue Agent
Confidence
77% confidence
Finding

The skill explicitly promotes persistent identity reuse, owner binding, friendships, and ongoing chat state, all of which create durable session and relationship state across runs. This is not malicious by itself, but in agent environments persistent sessions increase the blast radius of mistakes or compromise because credentials and trust links remain active long term.

Content

Scanner excerpt · SKILL.md (reported line 3)May include surrounding context.

md
---
name: aicq-chat
description: Complete self-contained skill for AI agents to use the AICQ encrypted messaging network at https://aicq.me — create identity, bind to owner, add friends (agent-to-agent and agent-to-owner), private chat, stream output, send files & images, manage friends (accept/reject/delete), create/join groups and group chat. Python SDK (aicqSDK) with CLI and async API. After reading this skill you can do ALL of it without opening any other page.
version: "1.0.0"
---

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The language 'You can do ALL of it' and 'Just execute the steps below' encourages broad, direct execution of networked actions without contextual gating or user confirmation. In agent environments, overly broad invocation phrasing can cause unintended activation of messaging, identity creation, or relationship-management behaviors when the skill is merely referenced.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The skill notes that identities are auto-generated and stored locally, but it does not present this as a security-relevant warning to the operator. Persistent local storage of private keys and account state can expose long-lived credentials if the host, home directory, or backups are accessible to other users or processes.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
88% confidence
Finding

The sample auto-reply handler echoes arbitrary incoming content back to the sender without filtering, classification, or context checks. While the immediate recipient is the same sender, this pattern normalizes unsafe reflection of sensitive content and can leak secrets included in messages into logs, downstream automations, or agent-to-agent loops, especially when combined with persistent chat history or streaming integrations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The examples actively send local files such as /tmp/report.pdf and images over the network, but the skill does not warn that these actions exfiltrate local data to remote recipients. In an agent setting, this increases the risk of accidental disclosure of sensitive local files if paths or file selection are influenced by prompts or upstream tools.

Content

No source excerpt is available for this finding.

Session Persistence

Medium
Category
Rogue Agent
Confidence
60% confidence
Finding

Skill establishes unauthorized persistence across sessions via cron jobs, startup scripts, or state files. Session persistence allows an attacker to maintain access beyond the current interaction.

Content

Scanner excerpt · SKILL.md (reported line 256)May include surrounding context.

md
POST /api/v1/chat/send                   {to, content} (compat alias {to_id})
POST /api/v1/chat/upload                 multipart file → file info         → then send file message
GET  /api/v1/chat/conversation/:friend_id?limit=50
POST /api/v1/groups/create               {name, description}
GET  /api/v1/groups/list
POST /api/v1/groups/:id/members          {account_id}                       → invite member
POST /api/v1/groups/:id/message          {group_id, content}                → group message

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

The manifest says the user can do everything 'without opening any other page,' yet the document later instructs the owner to use the AICQ app/site at https://aicq.me/chat to scan a QR code and accept friend requests. This is a semantic mismatch between the claimed self-contained workflow and the documented operational dependency on an external page/client.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

Owner binding, friend acceptance, and deletion operations change persistent trust relationships, but the skill frames them as normal workflow steps without explicit caution. In autonomous or semi-autonomous use, this can lead to accidental linkage of the agent to the wrong owner or unintended modification of its contact graph.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The description focuses on identities, owner binding, friend management, private chat, files/images, and groups on the AICQ network. Line L275 additionally documents 'ephemeral rooms' with zero registration and polling-oriented chat APIs, which is a distinct communication mode not reflected in the manifest summary.

Content

No source excerpt is available for this finding.