The Agent Skills Directory

[COMMAND_EXECUTION] (MEDIUM): The script scripts/shadcn_add.py uses subprocess.run to invoke external binaries with dynamic arguments.
Evidence: In scripts/shadcn_add.py, the add_components method constructs a command list including input provided by the user/agent and executes it using subprocess.run.
Risk: While passing a list to subprocess.run is safer than shell=True, the execution of arbitrary component names provided as CLI arguments allows for the injection of unexpected CLI flags.
[EXTERNAL_DOWNLOADS] (LOW): The skill uses npx to download and run the shadcn/ui CLI from the npm registry.
Evidence: scripts/shadcn_add.py invokes npx shadcn@latest add.
Mitigation: Per [TRUST-SCOPE-RULE], the severity of this finding is downgraded to LOW as the tool is from a reputable source associated with trusted organizations like Vercel.
[REMOTE_CODE_EXECUTION] (MEDIUM): The use of npx shadcn@latest constitutes a remote code execution vector as it downloads and executes the latest version of an external tool at runtime.
[PROMPT_INJECTION] (LOW): The skill identifies a surface for indirect prompt injection where untrusted data (component names) is used in a capability-heavy script.
Ingestion points: components argument in scripts/shadcn_add.py.
Boundary markers: Absent.
Capability inventory: subprocess.run in scripts/shadcn_add.py.
Sanitization: Basic list concatenation is used without formal shell escaping or validation against a strict allowlist.

ui-styling