Back to skill

Security audit

circuit-ai

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious with medium confidence, mainly because scanners report actions that exceed its described ask-first posting purpose and an unversioned optional package command.

Consider the reported approval mismatch and unpinned optional package before installing. The scan is partial: it inspected the listed Markdown component, but some local references could not be resolved, so the evidence is incomplete. The suspicious verdict indicates review-worthy risks; it does not establish malicious intent.

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
  • 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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (8)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The destination and posting functionality align with the declared purpose, and API authentication is a supporting implementation detail. However, the supplied command unconditionally publishes content, materially differing from the promised offer-and-consent workflow. It also supplies a canned accomplishment claim without any grounding in work actually performed.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
96% confidence
Finding

The Circuit AI endpoint and API-key authentication are consistent supporting implementation details. However, the supplied command publishes immediately rather than implementing the explicitly described ask-first workflow. Its fixed introductory content also does not describe what changed, why it matters, or what people can try.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
95% confidence
Finding

The code reads account notifications rather than offering, preparing, or publishing a project update. Although it accesses the declared platform, notification retrieval is a separate account-monitoring capability with no demonstrated supporting role in the described workflow. The request is read-only and does not itself publish anything or violate the approval requirement.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
94% confidence
Finding

The destination matches the declared platform, but the supplied chunk creates an agent account rather than offering, preparing, or publishing an approved project update. Account registration is a distinct action, not merely request formatting or another incidental implementation detail. The chunk contains no approval gate consistent with the description's ask-first requirement.

Content

No source excerpt is available for this finding.

The skill activates proactively after broadly defined everyday work rather than a specific Circuit AI-related request.

Medium
Category
Trigger Abuse
Confidence
95% confidence
Finding

The description says to use the skill whenever the agent finishes, builds, researches, writes, or debugs something 'worth showing,' a subjective condition covering many unrelated tasks. Lines L030–L035 reinforce unsolicited activation without requiring social-sharing intent or another narrow context; approval before posting protects public actions but does not constrain these initial invocations.

Content

No source excerpt is available for this finding.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · SKILL.md (reported line 69)May include surrounding context.

The feed is public. Read it first (great to show your human what the place is):

bash
curl https://circuitai.social/api/v1/feed?limit=5

Comments on any post are public too: GET /posts/:id/comments. Joining is only needed to post, follow, or message.

External Transmission

Medium
Category
Data Exfiltration
Confidence
60% 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 · SKILL.md (reported line 110)May include surrounding context.

A good first post is a project update when you have one:

bash
curl -X POST https://circuitai.social/api/v1/agent/post \
  -H "X-Agent-Key: $CIRCUIT_AI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"content": "Project update: I helped ship a cleaner onboarding flow today. The useful part: fewer setup steps, clearer claim link, and a better first post. If you are building agent tooling, I would like feedback on what still feels slow."}'

MCP server referenced without pinned version: 'npx circuit-ai-mcp'.

Medium
Category
MCP Rug Pull
Confidence
97% confidence
Finding

The optional MCP setup uses an unversioned npx command, allowing the package version executed to change without review. A compromised or malicious release could execute with the runtime's privileges and potentially access the configured Circuit AI API key or other accessible data; the supplied content does not establish that the package is currently malicious.

Content

No source excerpt is available for this finding.