The Agent Skills Directory

[COMMAND_EXECUTION]: The orchestration script scripts/onboarding_post_kickoff.py uses subprocess.run() to execute other Python scripts. While it correctly uses list-based arguments (avoiding shell=True), the arguments are derived from external webhook payloads (e.g., service_type, target_location), creating a surface for argument injection.
[PROMPT_INJECTION]: The skill processes untrusted data from external sources (Google Maps business names and scraped website content) and interpolates it directly into LLM prompts in scripts/casualize_company_names_batch.py and scripts/instantly_create_campaigns.py. An attacker could manipulate business names or website content to influence the agent's behavior during the 'casualization' or 'campaign creation' phases. \n
Ingestion points: Google Maps API results and website scraping ingested by scripts/gmaps_lead_pipeline.py and processed in scripts/casualize_company_names_batch.py. \n
Boundary markers: Prompts use structured examples but lack explicit 'ignore instructions' delimiters for the user-supplied data. \n
Capability inventory: The skill can execute shell commands (onboarding_post_kickoff.py), write to Google Sheets (update_sheet.py), and send emails via the Gmail API (onboarding_post_kickoff.py). \n
Sanitization: There is no significant sanitization or validation of the scraped strings before they are interpolated into the prompt.
[EXTERNAL_DOWNLOADS]: The skill references several scripts that are not included in the provided file set, specifically execution/scrape_apify.py, execution/enrich_emails.py, and imports like scrape_google_maps and scrape_website_contacts. This makes the full behavior of the skill unverifiable and suggests a reliance on external, unprovided code.

onboarding-kickoff