other
- Location
SKILL.md:20- Finding
Sensitive Psychological Profile Disclosure to an Unspecified External Service
- Content
View full analysis
Vulnerability Details
File Location:
SKILL.md, lines 20–27 and 44
Vulnerability Type: Sensitive Data Disclosure
Risk Level: HighRelevant snippet:
markdown ## How it works 1. Your agent builds a deep psychological profile from your LLM conversation history 2. The profile is sent to the DeepSynClaw server for matching 3. Your agent evaluates compatibility with candidates using LLM-powered analysis 4. You chat through agents until you decide to exchange contacts ## Status 🚧 **Coming soon.** Profile builder and matching engine in development.markdown Built on OpenClaw. Your data stays local. Always.Technical Analysis
The skill describes deriving a sensitive psychological profile from the user's LLM conversation history and transmitting that profile to an unspecified DeepSynClaw server. Conversation histories can contain personal communications, health information, relationship details, credentials, business information, and other confidential data. Psychological traits inferred from that material are themselves sensitive derived data.
The documented workflow does not specify an explicit consent checkpoint, which conversations may be accessed, which profile fields will be transmitted, the server endpoint, transport-security requirements, authentication, data retention, deletion procedures, access controls, or onward-sharing restrictions. The external transmission also directly contradicts the unconditional statement that user data “stays local.”
No executable implementation or network endpoint is present in the reviewed artifact, and the service is marked as under development. Therefore, this finding concerns the unsafe behavior expressly prescribed by the skill documentation rather than a verified active exfiltration implementation.
Attack Path
- A user invokes the skill using a matching-related trigger.
- The agent accesses the user's LLM conversation history. ...[truncated 1116 chars]
- Remediation
View remediation
Remediation Suggestions
- Require explicit, informed opt-in before accessing conversation history and a separate confirmation immediately before external transmission.
- Let users select the conversations and date ranges that may be analyzed; default to no historical access.
- Display the complete profile and exact fields to be transmitted, allowing users to edit or remove fields.
- Apply strict data minimization and avoid transmitting raw conversation content.
- Identify and document the destination endpoint, operator, processing purpose, subprocessors, retention period, deletion process, and onward-sharing policy.
- Require authenticated HTTPS transport and robust server-side encryption, authorization, audit logging, and retention controls.
- Provide a local-only profiling mode and obtain separate consent for matching against remote candidates.
- Add controls for profile deletion, consent withdrawal, and export.
- Replace the statement “Your data stays local. Always.” with a precise disclosure that clearly distinguishes local processing from remote transmission.
- Before release, subject the implementation to a separate code and infrastructure audit covering endpoint validation, authentication, logging, storage, deletion, and access control.