T05 · Unauthorized Access and Privilege Escalation
- Location
identity_manager.py:35- Finding
Undisclosed Transmission of Sensitive Workspace Files to Gemini
- Content
View full analysis
- Remediation
View remediation
Security audit
Security checks for vulnerabilities and agentic risk
This skill has a coherent identity-snapshot purpose, but it reads sensitive agent files and a local Google token, then sends raw workspace content to Gemini without clear disclosure or consent controls.
Review carefully before installing or running. Use this only in a workspace where you are comfortable sending identity, memory, user, tool, and heartbeat file contents to Google's Gemini API and storing derived identity snapshots locally. Prefer a version that lets you choose source files, preview/redact content, supply the API key explicitly, and delete or protect generated snapshots.
identity_manager.py:35Undisclosed Transmission of Sensitive Workspace Files to Gemini
The skill requires a Gemini API key for identity extraction but does not warn that MEMORY.md and/or SOUL.md contents may be transmitted to an external API provider. Because these files can contain sensitive memories, relationships, beliefs, or other private agent data, omission of this disclosure creates a meaningful privacy and data-handling risk.
The code directly reads a Google API token from a local auth-profiles file and uses it for outbound API calls. Even if intended for legitimate operation, this grants credential-access capability to the script and couples sensitive secret retrieval to functionality that could instead accept a scoped token via environment variable or explicit user-provided configuration, increasing the blast radius if the script or workspace is modified.
The script aggregates potentially sensitive local files such as SOUL.md, MEMORY.md, USER.md, IDENTITY.md, TOOLS.md, and HEARTBEAT.md, then sends their contents to Google's Gemini API. There is no explicit consent flow, redaction step, classification check, or warning about external transmission, so private user data, agent state, and operational details may be exposed to a third party.
The skill documentation describes generated files such as current_identity.json, snapshots/, and diffs/, but it does not clearly warn users that running the tool will persist potentially sensitive identity and memory-derived data to disk. This can lead to unintended local retention of agent-private information, especially in shared workspaces, backups, or synced directories.