rlama

SUSPICIOUS: the skill's document-indexing behavior fits its purpose, but its footprint is not fully consistent with the 'fully local/no data leaving the machine' claim because it supports cloud synthesis and arbitrary custom endpoints. The largest risk is supply-chain trust: the official RLAMA install uses raw GitHub curl|sh to fetch an unpinned binary with no visible integrity verification. Not confirmed malware, but medium-high risk and misleadingly framed as purely local.

This script deduplicates document files and rebuilds an RLAMA RAG, using filesystem deletion and external `rlama` subprocesses as part of its stated function. The command-line `rag_name` is not constrained to a safe name, so absolute paths or traversal components may redirect deletion outside `~/.rlama`; this is a significant destructive-path risk requiring validation and containment checks. No domains, URLs, IP addresses, credentials, or clear malware behavior appear in the fragment.