The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): The skill triggers npm install to set up dependencies like Vite and React. This action downloads third-party packages from the npm registry which can execute arbitrary code on the host machine during installation via preinstall/postinstall scripts.
COMMAND_EXECUTION (MEDIUM): The skill starts a Vite development server in the background. Running persistent background processes is a security concern as it can be used to hide malicious activity or maintain unauthorized access to local resources.
PROMPT_INJECTION (LOW): The skill reads and processes KANBAN.md, which serves as an indirect prompt injection surface. Maliciously crafted content within the Kanban file could attempt to influence the agent's behavior during the task management process.
Ingestion points: KANBAN.md (read/write access)
Boundary markers: Absent; the agent reads the file directly to manage cards.
Capability inventory: npm install, filesystem write (KANBAN.md, package.json), background process execution.
Sanitization: No sanitization of the markdown content is specified before processing.
DATA_EXFILTRATION (SAFE): The skill operates on localhost:5555 and local files. No evidence was found of sensitive data being transmitted to external, non-whitelisted domains.

kanban-app