The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied data, including notes, marking rubrics, and existing drafts, to generate or refine assignments. This creates a surface for indirect prompt injection where malicious instructions could be embedded in user content to influence the agent's behavior.
Ingestion points: Data enters the context through the Assignment Forge (user notes), the Draft stage (existing outlines/notes), and the Rewrite stage (user-pasted drafts).
Boundary markers: The instructions do not explicitly define strict boundary markers (e.g., delimiters) for separating user input from system instructions, relying on the model's contextual understanding.
Capability inventory: The skill can perform web searches using host tools, write project files to the project-memory/ directory, and generate long-form prose.
Sanitization: There is no explicit sanitization or filtering of user-provided content before it is interpolated into the planning or drafting prompts.
[EXTERNAL_DOWNLOADS]: The skill recommends the installation of companion tools from the author's GitHub repository to enhance research and writing capabilities.
Evidence: Instructions in SKILL.md and README.md suggest using npx skills add to download research-dojo and calligraph-sensei from https://github.com/kshanxs/skill-dojo. These references are to the author's own repository and are documented as standard extension of functionality.

truedraft