The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection within the generate_campaigns_with_claude function in scripts/instantly_create_campaigns.py.
Ingestion points: Untrusted data is ingested from external Google Sheets (via scripts/read_sheet.py), user-provided command-line arguments such as --client_description, and a local examples file .tmp/instantly_campaign_examples/campaigns.md.
Boundary markers: No delimiters or defensive instructions are used to isolate these variables when they are interpolated into the prompt for the Claude model.
Capability inventory: The agent has the capability to interact with the Instantly API to create email campaigns and the Anthropic API to generate content.
Sanitization: The skill does not sanitize or validate the content of the ingested data before prompt interpolation.
[EXTERNAL_DOWNLOADS]: The script scripts/read_sheet.py fetches data from Google Sheets, which is a well-known and trusted service from a recognized organization (Google). This data is used in downstream LLM prompts.
[COMMAND_EXECUTION]: The SKILL.md file contains instructions that involve executing local Python scripts with parameters derived from user-provided strings, which are passed as command-line arguments to the system shell.

instantly-campaigns