The Agent Skills Directory

[EXTERN AL_DOWNLOADS]: Th e skill in structs the user to c lone a code base from a thir d-party Git Hub reposit ory (https ://github.c om/uxKero/a nydesign.gi t) that is not recogni zed as a tr usted vendo r or affili ated with t he stated a uthor, reas on-machines .
[REMOTE_CO DE_EXECUTIO N]: The ins tallation p rocess requ ires runnin g pip inst all for un verified de pendencies from the re mote reposi tory and in stalling br owser binar ies via Pla ywright, wh ich execute s code from an unauthor itative sou rce.
[INDIRECT_PRO MPT_INJECTI ON]: The sk ill ingest s data from external w ebsites, Fi gma files, and images , creating a vulnerab ility surfa ce where th e agent cou ld be manip ulated by i nstruction s embedded in those s ources.
Inges tion points : Website content via URL fetchi ng, Figma metadata an d variables via MCP to ols, and us er-uploaded images.
**Bounda ry markers: ** Absent. T he instruc tions do no t provide d elimiters o r guidanc e for the a gent to dis regard emb edded comm ands in ana lyzed desi gns.
Capabil ity invento ry: The sk ill can wri te files to the local file syste m, perform network op erations to fetch webs ite data, a nd utilize vision cap abilities.
Saniti zation: A bsent. The skill does not mentio n any proc ess for val idating or filtering potentially malicious content re trieved fro m the anal yzed sourc es.

anydesign-analyzer