The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions direct the agent to download and execute a shell script from the official Hummingbot repository using bash process substitution: bash <(curl -s https://raw.githubusercontent.com/hummingbot/...). This pattern is used for both generating slides and editing them. While the source is the official vendor repository, this method involves executing remote code locally.
[EXTERNAL_DOWNLOADS]: The skill fetches its main execution logic (generate_slides.sh) from the vendor's GitHub repository. It also downloads and installs the fpdf2 Python package via pip3 if it is not already present on the system.
[COMMAND_EXECUTION]: The shell script and internal Python script execute several commands to perform their tasks: pip3 install fpdf2 to manage dependencies; mmdc (Mermaid CLI) or npx @mermaid-js/mermaid-cli via subprocess.run to render diagrams from text descriptions; mkdir -p to create output directories; and mktemp -d for temporary file management.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied markdown content to generate slides, creating an attack surface where malicious instructions could be embedded in the data.
Ingestion points: User-provided markdown files or text input (documented in SKILL.md).
Boundary markers: None identified in the prompt instructions to delineate user content from system instructions.
Capability inventory: File system writes (/tmp, ~/), network downloads (curl), and shell command execution (bash, mmdc, python3).
Sanitization: The Python script uses regular expressions to parse the markdown content before rendering it into the PDF document.

slides-generator