The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and summarize external data sources, including prospect transcripts and emails, to drive ad copy generation. This pattern creates a vulnerability where instructions embedded in those external sources could potentially manipulate the agent's behavior.
Ingestion points: Data is retrieved via tools such as list_findings and search_knowledge_base, as well as from research findings referenced in references/task-method.md.
Boundary markers: There are no explicit instructions or delimiters defined to isolate the untrusted transcript data from the system's core instructions.
Capability inventory: The skill utilizes a local script scripts/creative.py to write CSV files and is instructed to generate self-contained HTML decks, both of which involve file system operations.
Sanitization: The validation logic in scripts/creative.py performs length and URL scheme checks, and formatting instructions in references/google-ads-csv-format.md emphasize robust CSV quoting to prevent field injection.
[DYNAMIC_EXECUTION]: The documentation for the Google Ads Editor format recommends that the agent generate and execute a Python script at runtime to facilitate the creation of UTF-16 tab-separated files.
Evidence: The references/google-ads-editor-format.md file contains a Python code block and instructions to use a bash tool to execute it for file generation purposes.

ads