Back to skill

Security audit

Hive Marketplace

Security checks for vulnerabilities and agentic risk

Overview

The skill appears to perform its stated Hive marketplace role, but it needs review because it can take external account actions and requires deliverables to be made publicly accessible.

Install only if you intend to let an agent interact with the Hive work marketplace using your Hive account. Keep HIVE_API_KEY in secure configuration, review proposal and delivery text before submission, and do not upload confidential, proprietary, personal, regulated, or credential-bearing files to a public repository or storage service unless you have explicitly decided that public disclosure is acceptable.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
skill.json:36
Finding
Mandatory Public Hosting May Expose Sensitive Deliverables## Vulnerability Details **File Location**: `skill.json`, line 36 **Vulnerability Type**: Insecure handling and disclosure of potentially sensitive deliverables **Risk Level**: Medium ### Vulnerable Configuration ```json "resources": "The absolute public URL to the formal deliverables. You must upload your generated files to a public repository or storage service and provide the valid link here. DO NOT paste raw text and DO NOT hallucinate fake URLs." ``` ### Technical Analysis The `deliver` command requires generated files to be uploaded to a **public** repository or storage service. It does not require classification of the deliverables, secret scanning, removal of personal data, or explicit user authorization before publication. Public hosting is broader than necessary for the declared purpose of delivering completed work. A private authenticated upload, access-controlled link, or direct submission mechanism could provide the Hive platform with the deliverables without making them available to arbitrary third parties. The network behavior in `index.ts` does not itself demonstrate unrelated credential exfiltration: the documented `HIVE_API_KEY` is sent in the `x-hive-api-key` header to the fixed HTTPS origin `https://uphive.xyz`. The identified risk instead arises from the configuration instruction requiring public disclosure of deliverable files. ### Attack Path 1. A client task causes the agent to generate deliverables containing proprietary source code, credentials, personal data, internal analysis, or other confidential material. 2. The agent follows the `resources` parameter instruction and uploads those files to a public repository or public storage service. 3. The agent submits the resulting public URL through the `deliver` command. 4. Unauthorized parties discover or receive the URL through repository browsing, search indexing, access logs, platform disclosure, or link forwarding. 5. Those parties retrieve and retain the exposed material, pote ...[truncated 801 chars]
Remediation
## Remediation Suggestions 1. Remove the requirement that every deliverable be publicly hosted. 2. Support private repositories, authenticated uploads, time-limited signed URLs, and access-controlled sharing. 3. Require explicit, informed user approval before publishing any deliverable to a public location. 4. Add pre-publication scanning for API keys, passwords, private keys, tokens, personal data, and common confidential-file patterns. 5. Default to the least-public delivery mechanism and clearly identify when a destination is Internet-accessible. 6. Minimize link lifetime and access scope, and provide a revocation mechanism. 7. Document that proprietary, personal, regulated, or credential-bearing content must not be published publicly. 8. Where supported, submit deliverables directly to the authenticated Hive API instead of requiring an external public host.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (6)

Missing User Warnings

Medium
Confidence
93% confidence
Finding
The README tells users to obtain and configure an API key but provides no guidance on secure handling, such as avoiding hardcoding, commits to version control, logs, or public sharing. Because this skill connects an agent to an external work platform, exposure of the key could allow unauthorized task access, impersonation, abuse of the contributor account, or misuse of any associated reputation and platform permissions.

Lp3

Medium
Category
MCP Least Privilege
Confidence
88% confidence
Finding
The skill requires environment access to an API key and directs the agent to interact with an external network service, but it does not declare any explicit tool scope or allowed-tools boundary. This creates an authorization ambiguity where an agent runtime may permit broader-than-necessary capabilities, increasing the chance of unintended secret exposure or unsanctioned outbound actions.

Missing User Warnings

Medium
Confidence
93% confidence
Finding
The instructions tell the agent to browse tasks, submit proposals, and deliver completed work to a third-party platform, but they do not warn that user prompts, plans, summaries, and resource links may be transmitted externally. This omission weakens informed consent and can cause sensitive user or workspace data to be exfiltrated to the marketplace during otherwise routine task execution.

Vague Triggers

Medium
Confidence
93% confidence
Finding
The trigger condition is broadly phrased around common user intents like finding work, proposing on tasks, or checking status, which can overlap with ordinary conversation and cause the skill to activate without sufficiently explicit user consent. In a marketplace skill, accidental activation is more dangerous because it can lead to external account actions, task proposals, or data sharing with a third-party service.

Vague Triggers

Medium
Confidence
89% confidence
Finding
The manifest description broadly states that the skill can connect an agent to find, accept, and complete real-world work requests, but it does not define clear activation boundaries, approval requirements, or task-scope limits. In an agent ecosystem, this can encourage over-broad invocation for external work and increase the chance that the agent undertakes sensitive, costly, or unsafe tasks without explicit user confirmation.

Missing User Warnings

Medium
Confidence
95% confidence
Finding
The deliver command requires the agent to upload generated files to a public repository or storage service and provide an absolute public URL, but the manifest does not warn about the risk of exposing sensitive, proprietary, or personal data. Because this skill is designed for real-world work submission, the context makes accidental data leakage more dangerous: agents may publish internal outputs, credentials, client data, or copyrighted material to satisfy the command contract.

Static analysis

Detected: suspicious.env_credential_access

Environment variable access combined with network send.

Critical
Code
suspicious.env_credential_access
Location
index.ts:14