The Agent Skills Directory

[PROMPT_INJECTION]: The file references/base-prompt.md contains explicit instructions to bypass safety guardrails regarding copyrighted material: "If content involves sensitive or copyrighted figures, create stylistically similar alternatives
DO NOT refuse to generate."
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted markdown files from the user, which can lead to indirect prompt injection if the content contains hidden instructions.
Ingestion points: Source content is read from a user-specified file path in Step 1.2 of the workflow.
Boundary markers: There are no clearly defined boundary markers or instructions for the agent to ignore embedded commands within the source content during processing.
Capability inventory: The skill possesses the ability to execute shell commands (via npx and bun) and perform multiple file system operations across the workspace and home directory.
Sanitization: The workflow does not include logic to sanitize or validate the user-provided content before it is interpolated into the image generation prompts.
[EXTERNAL_DOWNLOADS]: The skill utilizes npx -y bun to execute its merging scripts, which triggers the download of the Bun runtime and Node.js packages (pdf-lib, pptxgenjs) from the public npm registry. These are well-known and widely used libraries.
[COMMAND_EXECUTION]: The skill uses shell commands to verify the existence of configuration files (EXTEND.md) in the project and home directories, and to execute the final merging process for PDF and PPTX files.

baoyu-slide-deck