忆时

• Analyzed: May 3, 2026, 8:46 AM
• Findings: 1
• Version: 1.0.0
• Hash: 5f9cf1b83bed0b78c9f48df6979b61fc5542b94c73e97f552d51ce246932b71d
• Source: palm
• External report: View on VirusTotal

Findings (1)

The '忆时' (Memocap) skill implements a local memory management system using ChromaDB but exhibits several high-risk characteristics. The `memory_core.py` script provides the agent with unconstrained file read and write capabilities through its import and export functions without path sanitization, which could be exploited to exfiltrate sensitive files (e.g., SSH keys or configuration files). The script also employs an OS-level stderr redirection (`_silent_import`) to suppress system messages, a technique that can be used to mask execution errors or malicious activity. Furthermore, the `yishi-instructions.md` file serves as a prescriptive prompt injection that globally alters the agent's persona and mandates the execution of shell commands during every interaction, significantly increasing the attack surface. The presence of hardcoded absolute paths (e.g., `/home/fslong/`) in `SKILL.md` and `modules/01-initialize.md` further indicates that the bundle is untrusted and potentially tailored for a specific environment.