The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads untrusted data (first names, company names, cities) from an external Google Sheet and interpolates this content directly into prompts sent to the Anthropic API. If the spreadsheet contains malicious instructions, they could influence the model's behavior.
Ingestion points: Data is read from user-provided spreadsheets via worksheet.get_all_values() in scripts/casualize_batch.py, scripts/casualize_city_names_batch.py, scripts/casualize_company_names_batch.py, and scripts/casualize_first_names_batch.py.
Boundary markers: Absent. The data is placed directly into the prompt string without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill has the ability to communicate with the Anthropic API for text generation and perform read/write operations on the user's Google Sheets via the Google Sheets API.
Sanitization: Absent. The scripts do not perform any validation or escaping of the spreadsheet content before processing.
[COMMAND_EXECUTION]: The skill's documentation includes standard shell commands for running the provided Python scripts. These commands are typical for local execution of the skill's functionality.
Evidence: SKILL.md provides examples like python3 -u ./scripts/casualize_batch.py "GOOGLE_SHEET_URL" for the user to execute.

casualize-names