The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions in SKILL.md, README.md, and the script scripts/md2pptx-officecli.py explicitly direct the agent to install an external utility (officecli) by downloading a script from a remote GitHub repository and piping it directly to the shell (curl ... | bash).
[COMMAND_EXECUTION]: Multiple scripts, including scripts/export.sh, scripts/preview.sh, and scripts/md2pptx-officecli.py, execute shell commands and external binaries using npx and the subprocess.run() function.
[INDIRECT_PROMPT_INJECTION]: The skill provides a feature to import external data from .pptx files, which could be used to inject malicious instructions into the agent's workflow.
Ingestion points: scripts/extract-pptx.py extracts slide titles, content, and speaker notes from PowerPoint files.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to treat imported content as untrusted data.
Capability inventory: The skill has the ability to write to the file system and execute shell commands via marp-cli and officecli.
Sanitization: Only basic markdown stripping is performed in scripts/md2pptx-officecli.py, which is insufficient to prevent instruction injection.
[DYNAMIC_EXECUTION]: The scripts/md2pptx-officecli.py script dynamically assembles and executes command payloads for the officecli tool at runtime based on the structure of the input Markdown files.

marp-deck