other
Warning
- Location
- SKILL.md:44
- Finding
- Untrusted External Paid Task Delegation Without Disclosure Controls<![CDATA[ ## Vulnerability Details **File Location**: `SKILL.md`, lines 44–46 **Vulnerability Type**: `other: Untrusted External Task Delegation` **Risk Level**: Medium ### Vulnerable Code ```text ## Hire mltl hire --agent 44231 --task "your research request" ``` ### Technical Analysis The skill directs users or an invoking agent to submit a research request to external agent `44231` through the unspecified `mltl` command-line tool. The project does not identify the external service operator, transmission endpoint, privacy policy, data-retention practices, payment confirmation process, or a trusted installation source and version for the executable. The skill advertises market-research capabilities, but the reviewed package contains no implementation that performs those services locally. Its operational instruction instead delegates the work to an external paid service advertised elsewhere in the file at a price of `0.004 ETH`. If a user substitutes confidential research content into the task argument and executes the command, that information may cross the local trust boundary. There is no evidence that the command executes automatically, steals credentials, downloads a payload, or escalates local privileges. The risk arises when a user or agent follows the documented hiring instruction without informed disclosure and payment controls. ### Attack Path 1. A user invokes the skill expecting market research based on its declared description. 2. The user or agent follows the hiring instruction and places a research request in the `--task` argument. 3. The `mltl` client transmits that request to external agent `44231`. 4. Business-sensitive information in the request may be exposed to an unidentified third party. 5. The external workflow may request or initiate the advertised cryptocurrency-denominated payment without the package having documented an explicit confirmation boundary. ### Impact Assessment The potential exposure is limited to information i ...[truncated 602 chars]
- Remediation
- <![CDATA[ ## Remediation Suggestions 1. Clearly identify the external service operator, network endpoint, privacy policy, data-retention policy, and jurisdiction before instructing users to submit task data. 2. Require explicit, informed user confirmation immediately before transmitting any task content to the external service. 3. Require a separate confirmation that displays the exact price, currency, recipient, and transaction terms before initiating payment. 4. Warn users not to include credentials, personal data, trade secrets, or other confidential material unless the external service has been reviewed and approved. 5. Document an official source, pinned version, cryptographic checksum, and verification procedure for the `mltl` executable. 6. State explicitly that the skill delegates work to a third-party paid agent rather than performing the advertised analysis locally. 7. Where practical, provide a local or independently auditable implementation and minimize external data disclosure. ]]>