The Agent Skills Directory

[PROMPT_INJECTION]: The scripts/generate_proposal.py script is susceptible to indirect prompt injection because it ingests untrusted meeting transcripts and incorporates them directly into the agent's prompt context.
Ingestion points: Meeting transcripts provided via the --transcript or --transcript-file arguments in scripts/generate_proposal.py.
Boundary markers: No delimiters or unique separators are used to isolate the untrusted transcript from the system instructions.
Capability inventory: The script can create Google Documents and perform autonomous web research via the parallel_research module.
Sanitization: There is no evidence of input validation, escaping, or filtering applied to the transcript content before processing.
[COMMAND_EXECUTION]: Several scripts utilize system commands to perform their primary functions.
scripts/md_to_pdf.py calls the Chrome/Chromium executable in headless mode to convert HTML to PDF.
scripts/generate_flowchart.py calls the Mermaid CLI (mmdc) to render diagrams.
[EXTERNAL_DOWNLOADS]: The skill requires several external dependencies and tools.
Fetches and uses Python libraries such as pydrive2, pydantic-ai, and google-api-python-client.
Documentation recommends installing the Mermaid CLI via npm to enable diagram rendering.
Attempts to import a local module parallel_research for web-enabled research capabilities.

content-generation