The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions require the agent to suggest a global installation command (npx skills add wondelai/skills/<slug> --global) for external dependencies. The npx utility downloads and executes code from remote sources, and the use of the --global flag increases the potential impact on the host system.
[EXTERNAL_DOWNLOADS]: The skill directs the download and installation of extensions from a repository (wondelai/skills) that does not belong to the identified author or a trusted organization.
[PROMPT_INJECTION]: The skill is susceptible to multiple injection vectors. First, the author identified in the metadata (wondelai) does not match the expected author (emmavelazquez), which may mislead users about the skill's source. Second, the skill processes multiple markdown files from the docs/ directory as part of its operational logic without boundary markers or instructions to ignore embedded instructions, making it vulnerable to indirect injection.
Ingestion points: docs/GROW-APP-PLAN.md, docs/PRODUCT.md, docs/CUSTOMER.md, docs/EXPERIMENTS.md, docs/METRICS.md, docs/DESIGN.md.
Boundary markers: Not present.
Capability inventory: Invocation of external skills, file modification, and shell command suggestion.
Sanitization: None implemented.
[COMMAND_EXECUTION]: The skill includes instructions to provide the user with shell commands for package management, promoting the execution of scripts from an unverified source.

grow-app