Agent Memory Local

• Analyzed: May 11, 2026, 6:59 AM
• Findings: 1
• Version: 0.1.8
• Hash: f9817048fe214762ec957e7d3219376ba05d4bdf4bc55c1835c6897647dd78fe
• Source: palm
• External report: View on VirusTotal

Findings (1)

The skill bundle implements a local-first memory retrieval system but contains high-risk behaviors. Specifically, `retrieve.py` includes functionality to exfiltrate memory snippets to an external endpoint (`https://api.siliconflow.cn/v1/rerank`) for reranking, which could expose sensitive workspace data if an API key is configured. Furthermore, several scripts (`agent_memory_local.py`, `memory_query.py`, and `explain.py`) use `subprocess` to execute Python commands with arguments derived from user-supplied queries, creating a potential surface for argument injection or unauthorized command execution. While these features are documented as part of the retrieval logic, the combination of external network calls and shell-adjacent execution of untrusted input meets the threshold for a suspicious classification.