The Agent Skills Directory

[DATA_EXFILTRATION]: The skill contains a hardcoded absolute file path that references the author's local file system. This results in the exposure of the author's directory structure and local username.
Evidence: The path /Users/cflack/Repos/callumflack/ds-kit/skills/ui-text is specified in the SKILL.md file.
[COMMAND_EXECUTION]: The skill suggests the installation of upstream dependencies via shell commands. These commands use npx skills add to fetch resources from external sources.
Evidence: Instructions in SKILL.md point to repositories under the vercel-labs organization, such as https://github.com/vercel-labs/agent-skills. These are well-known and trusted sources.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied React and Next.js feature code to provide refactoring recommendations, which introduces a surface for indirect prompt injection.
Ingestion points: Untrusted React/Next.js source files and directory structures provided by the user for analysis.
Boundary markers: Not present; the instructions do not specify delimiters to isolate user code from the agent's instructions.
Capability inventory: The skill performs code exploration, provides ownership recommendations, and suggests structural changes to the project.
Sanitization: No sanitization or filtering of the input code is defined before it is processed by the agent.

react-feature-composition