The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and evaluate external data from GitHub repository descriptions and topics. This introduces a surface for indirect prompt injection where an attacker could place malicious instructions within a repository's metadata to attempt to subvert the agent's analysis.
Ingestion points: The ScanGitHubRepo function in SKILL.md fetches data from the api.github.com endpoint.
Boundary markers: There are no explicit instructions or delimiters in the provided code logic to differentiate between analyzed metadata and system instructions.
Capability inventory: The skill's implementation logic includes network access via Go's http package and environment variable access for authentication.
Sanitization: The skill parses JSON responses but does not sanitize the text content of repository descriptions or topics before they are processed for keyword matching.
[SAFE]: The skill accesses the GITHUB_TOKEN environment variable to authenticate requests to the official GitHub API. This is a standard method for handling credentials in development tools and the token is sent only to the legitimate service provider.
[SAFE]: All external URLs for reporting abuse and downloading legitimate software point to well-known technology companies and security vendors, such as Microsoft, Bitdefender, and Malwarebytes.

security-awareness-malicious-repository-detection