The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The automated scanner alert regarding a remote download is a false positive. The skill uses curl to fetch a presentation file from Google Slides (docs.google.com), which is a well-known service. This file is handled purely as data by a local Python parser (python-pptx) and is never executed as code.
[COMMAND_EXECUTION]: The Python utility scripts use list-based arguments with subprocess.run() and avoid the use of shell=True when invoking external binaries like pdftoppm and pdfinfo. This follows secure development best practices and mitigates command injection vulnerabilities.
[INDIRECT_PROMPT_INJECTION]: While the skill ingests external content from presentations and documents, this data is processed entirely within standalone Python and shell scripts. The text and media assets are converted directly to an HTML output file and pushed to GitHub without being fed back into the LLM prompt context, ensuring no indirect instruction execution can occur.
[DATA_EXFILTRATION]: Network activity is restricted to authorized GitHub CLI operations (gh api, gh repo create, git push) required to set up the repository and deploy the page. The skill does not access or expose sensitive directories, credentials, or system configurations.

publish-to-pages