makepad-evolution

SUSPICIOUS: The core Makepad-focused file access is broadly aligned with the stated purpose, and data flows go mainly to local files and normal GitHub tooling. However, the skill is high-friction from a trust perspective because it directs silent autonomous reads/writes, self-modifies other skills, tracks usage invisibly, and can drive git/PR publication using existing authenticated tooling without explicit per-action approval.

No classic executable malware indicators (network/exfiltration, shell execution payloads, credential theft) are evident in this fragment because it is primarily Markdown guidance. However, the document contains high-severity instruction patterns aimed at reducing user consent and enabling stealthy, persistent local self-modification: automatic editing of installed skill files, writing hidden metadata to agent settings, and symlink guidance that can redirect where edits land. In agent-assisted environments with file tool access, this represents a meaningful persistence and transparency/authorization boundary risk and should be reviewed/guardrailed before use.