The Agent Skills Directory

[COMMAND_EXECUTION]: The Bash script example in SKILL.md uses the eval command to execute a curl request containing variables ${REPO_OWNER} and ${REPO_NAME}. If these variables are derived from untrusted user input and not strictly sanitized, an attacker could inject shell commands (e.g., by providing a repository name containing shell metacharacters like ; or `).
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting and returning data from external GitHub contributor profiles.
Ingestion points: The skill fetches data from the GitHub API via the /contributors endpoint.
Boundary markers: None identified; the agent is prompted to return the login and name fields directly as JSON.
Capability inventory: The skill provides logic for network requests (curl, fetch) and demonstrates shell execution capabilities.
Sanitization: No sanitization or validation of the fetched contributor strings is performed before they are returned to the agent context.
[EXTERNAL_DOWNLOADS]: Fetches contributor metadata from the official GitHub REST API.

random-contributor