The Agent Skills Directory

[PROMPT_INJECTION]: The references/base-prompt.md file contains steering instructions designed to bypass the safety filters of downstream image generation tools. It explicitly instructs the AI to generate alternatives rather than refusing to generate when sensitive or copyrighted figures are mentioned ('DO NOT refuse to generate').
[INDIRECT_PROMPT_INJECTION]: The skill presents an indirect prompt injection attack surface by ingesting and processing untrusted external content.
Ingestion points: The agent is instructed to read source material from files or URLs provided by the user (Step 1.2 in references/workflow.md).
Boundary markers: The references/base-prompt.md uses structured sections like STYLE_INSTRUCTIONS and SLIDE CONTENT to delimit content, which can help prevent accidental obedience but does not fully mitigate adversarial input.
Capability inventory: The skill uses subprocesses to run Node.js and Python scripts for image manipulation and document merging, and it invokes an external image generation skill.
Sanitization: There is no evidence of explicit sanitization or filtering of the user-provided text before it is interpolated into the prompts used for image generation.
[COMMAND_EXECUTION]: The skill workflow requires the agent to execute included scripts (scripts/merge-to-pptx.ts, scripts/merge-to-pdf.ts, and scripts/apply-branding.py) using npx, bun, and python3. While these scripts are part of the skill package and perform intended tasks like compositing logos and merging files, they represent the execution of local code via command-line tools.

lov-any2deck