The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands using the npx skills CLI. Specifically, it encourages the use of the -y flag (npx skills add <package> -g -y), which bypasses confirmation prompts when installing new software, potentially leading to the unintended installation of malicious code if the agent is manipulated.
[EXTERNAL_DOWNLOADS]: The skill's primary function is to facilitate the download and installation of external packages from the open agent skills ecosystem. Although it mentions trusted sources like Vercel and Anthropic, it also enables installation from any GitHub repository or source provided through the search mechanism.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data (skill search results, descriptions, and metadata) which could be used as a vector for indirect prompt injection. A malicious skill listed in the registry could contain instructions in its metadata that attempt to influence the agent's behavior when discovered.
Ingestion points: Data enters the context via the output of npx skills find [query] and search results from skills.sh (SKILL.md).
Boundary markers: The instructions lack explicit boundary markers or delimiters for handling the untrusted text from search results.
Capability inventory: The skill has the capability to write to the file system and install global packages via npx skills add (SKILL.md).
Sanitization: The skill includes human-readable instructions for the agent to verify reputation (install count, stars), which acts as a heuristic filter but not a technical sanitization mechanism.

find-skills