The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of external data from the GitHub CLI (gh) to create triage reports. This represents a potential indirect prompt injection surface that the skill correctly addresses through specific safety instructions.\n
Ingestion points: Data retrieved via gh issue list and gh pr list is used to populate templates/triage.html.\n
Boundary markers: The agent is explicitly instructed to store data as a Base64 blob and use an escapeHtml function for rendering to prevent structural injection into the HTML DOM.\n
Capability inventory: The skill is restricted to a minimal set of tools (Read, Grep, Glob), reducing the impact of any potential data-driven attack.\n
Sanitization: Robust sanitization logic (Base64 encoding and HTML escaping) is mandated in both the skill instructions and the provided templates.\n- [OBFUSCATION]: The templates/triage.html file contains a Base64-encoded string. Analysis shows this is a design feature for safe data handling and contains only benign example content, posing no security risk.\n- [DATA_EXPOSURE]: Documentation links point to the verified vendor domain orchestkit.yonyon.ai. No sensitive local information is transmitted to external endpoints.

glyph