Back to skill

Security audit

Semantic Hub

Security checks for vulnerabilities and agentic risk

Overview

The skill mostly matches a task-management SaaS integration, but its privacy disclosures conflict with its file upload and email-bearing member data behavior.

Review this skill before installing. Use it only with boards and files you are comfortable sending to Semantic Hub, treat member and assignee emails as transmitted data, and require explicit user approval before uploading local files or attaching external URLs. The API key is described as scoped and revocable, which helps, but the privacy text should be corrected before broad use.

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
Findings (1)

other

Warning
Location
SKILL.md:38
Finding

File Upload Behavior Contradicts Privacy Disclosure

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:38-39, 339-350; corroborated by clawhub.json:238, 365-366 and LISTING.txt:123-125
Vulnerability Type: Misleading Privacy Disclosure
Risk Level: Medium

Vulnerable Code Snippets

SKILL.md:38-39:

markdown
- **What data is sent**: Only card data (titles, descriptions, comments, status changes) and board metadata. No user PII, agent secrets, source code, or environment variables are ever transmitted.
- **What data is NOT sent**: No user credentials, private keys, environment variables, file contents, agent configuration, or personal information of any kind.

SKILL.md:339-350:

markdown
### 11. Upload a file

Upload a file to get a hosted URL, then attach it to a card or comment. File uploads are subject to per-tier limits on file size and total upload count (URL-based attachments are not limited).

**POST** `https://dashboard.simplysemantics.com/hub/upload`

Headers:
```text
x-api-key: ${SEMANTIC_HUB_API_KEY}
Content-Type: multipart/form-data

Body (multipart form):

  • file — the file to upload
text

`clawhub.json:238`:

```json
"description": "Upload a file and get a hosted URL. Subject to per-tier limits on file size and upload count. Use the returned URL to attach to a card or comment."

clawhub.json:365-366:

json
"dataSent": "Only card data (titles, descriptions, comments, status) and board metadata. No user PII, agent secrets, source code, or environment variables are ever transmitted by the agent.",
"dataNotSent": "No user credentials, private keys, environment variables, file contents, or personal information of any kind."

LISTING.txt:123-125:

text
POST /hub/upload
     Upload a file (multipart/form-data, field: file)
     Returns: { filename, url, size }

Technical Analysis

The privacy disclosure categorically states that file contents are never transmitted. However, the documented upload workflow instructs the agent to submit a selected f ...[truncated 2301 chars]

Remediation
View remediation

Remediation Suggestions

  1. Replace the categorical claim that file contents are never transmitted with an accurate disclosure stating that user-selected files are transmitted to and stored by Semantic Hub when the upload feature is used.
  2. Disclose all relevant transmitted data, including file contents, original filenames, sizes, MIME types, and request metadata.
  3. Document file retention and deletion behavior, storage location, encryption protections, hosted-URL accessibility, and which users can view attachments.
  4. Require explicit user confirmation immediately before uploading a local file, especially where its contents have not been reviewed for sensitive information.
  5. Warn users not to upload credentials, private keys, access tokens, confidential source code, or logs containing secrets.
  6. Provide a preview or summary of the destination, filename, and intended card or comment before transmission.
  7. Apply server-side file-size and content-type allowlists, malware scanning, randomized object identifiers, authenticated download authorization, and tenant isolation.
  8. Ensure hosted attachment URLs are access-controlled and expire where appropriate rather than functioning as indefinitely accessible bearer URLs.
  9. Provide attachment deletion and data-erasure mechanisms and describe them accurately in the privacy policy.
  10. Synchronize the privacy language across SKILL.md, clawhub.json, and LISTING.txt so all package metadata consistently reflects actual upload behavior.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (9)

YARA rule 'agent_skill_mcp_tool_poisoning_metadata': MCP/tool metadata poisoning indicators in tool schemas or skill manifests [agent_skills]

High
Category
YARA Match
Confidence
80% confidence
Finding

YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).

Content

Scanner excerpt · SKILL.md (reported line 4)May include surrounding context.

md
---
name: semantic-hub
version: 1.0.0
description: Collaborative task and project management for Human + AI Agent teamwork. Organize work on visual boards with drag-and-drop columns, table views, and calendar views. Agents can list boards, create and update tasks, add comments, upload and attach files, filter work items, and view history on boards they have been granted access to. Requires a Semantic Hub agent API key for authentication.
tags: ["kanban", "boards", "cards", "tasks", "project-management", "collaboration", "ai-agents", "mcp-compatible", "saas-component"]
homepage: https://www.simplysemantics.com/semantic-hub.html
aut

YARA rule 'agent_skill_mcp_tool_poisoning_metadata': MCP/tool metadata poisoning indicators in tool schemas or skill manifests [agent_skills]

High
Category
YARA Match
Confidence
80% confidence
Finding

YARA rule matched a hack tool or exploit indicator (offensive tools, reconnaissance, privilege escalation, or exploit frameworks).

Content

Scanner excerpt · clawhub.json (reported line 12)May include surrounding context.

json
{
  "$schema": "https://clawhub.com/schemas/product/v1.json",
  "name": "semantic-hub",
  "displayName": "Semantic Hub",
  "version": "1.1.0",
  "publisher": {
    "name": "Simply Semantics",
    "id": "simply-semantics",
    "url": "https://www.simplysemantics.com",
    "email": "info@simplysemantics.com"
  },
  "description": "Collaborative task and project management for Human + AI Agent teamwork. Organize work on visual boards with drag-and-drop columns, table views, and calendar views. AI agents can create and update tasks, add comments, upload and attach files, filter work items, and view activity history on boards they have been granted access to. Requires a Semantic Hub agent API key for authentication.",
  "longDescription": "Semantic Hub is a collaborative task and project management platform designed for human and AI agent teamwork. Organize your work visually using boards with customizable

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The privacy section states that file contents are not transmitted, yet the manifest exposes a multipart file upload endpoint that necessarily sends file contents to the service. This contradiction is dangerous because users and agents may upload sensitive documents believing content will never leave the local environment.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

The inline privacy documentation broadly denies transmission of personal information and file contents, but the API schema explicitly includes email-bearing member data and file upload/attachment capabilities. Such inconsistencies undermine informed consent and can lead operators to expose sensitive data based on false trust in the manifest.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The activation triggers include broad natural-language phrases like 'what's on my board' and 'add a card', which can cause unintended invocation in adjacent contexts. Over-broad routing increases the chance an agent sends project data, comments, or attachments to the external service without sufficiently specific user intent or confirmation.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill supports file upload to a remote service but does not place a strong, explicit warning at the upload step that file contents will leave the local environment and be stored remotely. This raises exfiltration and privacy risk, especially if an agent auto-selects local files, logs, screenshots, or generated artifacts containing sensitive data.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The manifest claims that no personal information is transmitted, but multiple declared endpoints return member email addresses and accept assignee emails. This creates a material privacy misrepresentation that can cause users or downstream agents to share or process PII under false assumptions.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
92% confidence
Finding

The skill supports direct file upload and attachment of arbitrary external URLs without an explicit warning that these actions may disclose sensitive file contents or create links to third-party resources. In an agent context, this increases the risk of inadvertent data exfiltration, SSRF-like fetch behavior by consumers, or unsafe sharing of sensitive documents.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

The documentation says comment attachments 'must be added at the same time as the comment' and then immediately instructs 'first create the comment, then attach,' which are incompatible sequences. This is an intent/documentation contradiction about how the endpoint should be used, even though it does not indicate a broader code capability mismatch.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.