The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes CSV files containing player-submitted comments and ratings, which serves as a potential ingestion surface for untrusted data.
Ingestion points: As noted in references/analysis.md, the skill reads CSV exports containing player ratings and free-text comments.
Boundary markers: The instructions in references/html-readout.md explicitly direct the agent to treat CSV values as "untrusted content" and use appropriate delimiters.
Capability inventory: The skill enables data record counting, synthesis of themes, and the generation of self-contained HTML reports.
Sanitization: references/html-readout.md requires the agent to render all comments and values as escaped text, ensuring they are not interpreted as markup or instructions. It also mandates the removal of sensitive player addresses from all report outputs.
[EXTERNAL_DOWNLOADS]: The skill references and directs the download of modules, assets, and documentation from the vendor's official repositories and recognized academic/professional entities.
Evidence: The skill links to official Decentraland repositories (e.g., github.com/decentraland/sdk7-test-scenes) for installation and configuration guidance.
References: Includes citations to ethical survey guidelines from the American Association for Public Opinion Research (AAPOR) and scholarly research on incentives via standard DOI links.

scene-feedback