The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from GitHub search results (repository names and descriptions) and arXiv paper titles, which are then interpolated directly into a prompt template for the 'Prepare Agent' using placeholders like {github_result} and {references}. An attacker could potentially influence the Prepare Agent's behavior by crafting malicious repository descriptions.
Ingestion points: External data from GitHub Search API results and arXiv metadata.
Boundary markers: The prompt templates lack explicit delimiters or instructions to ignore embedded commands within the interpolated variables.
Capability inventory: The skill uses write_file to save logs and paper sources and performs network operations via the GitHub and arXiv APIs.
Sanitization: There is no evidence of sanitization or filtering of the repository descriptions before they are added to the agent's context.
[DYNAMIC_EXECUTION]: The skill instructions specify that it should import a 'metaprompt module' based on a provided category parameter. Dynamic loading of code or configuration from variable paths or parameters can be a risk if the mapping logic is not strictly limited to a predefined whitelist.
[EXTERNAL_DOWNLOADS]: The skill fetches research materials from well-known services, including repository metadata via the GitHub API and LaTeX source archives from arXiv. These operations target trusted domains and are documented as part of the intended research workflow.

inno-prepare-resources