The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Automated alerts for curl piped to python3 in SKILL.md were evaluated. The commands utilize Python to parse and format public JSON data from https://status.customer.io/api/v2/status.json and are confirmed to be benign diagnostic tools rather than malicious remote code execution.
[CREDENTIALS_UNSAFE]: The skill references Customer.io API credentials. These are managed through environment variables such as CUSTOMERIO_TRACK_API_KEY, which aligns with standard security practices for secret management.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Customer.io API responses, creating an attack surface for indirect prompt injection.
Ingestion points: API responses for customer profiles, messages, and segments are fetched in SKILL.md and references/implementation.md.
Boundary markers: The skill lacks explicit delimiters or instructions telling the agent to treat external API data as untrusted or to ignore instructions embedded within that data.
Capability inventory: The skill utilizes Bash, Write, and Edit tools.
Sanitization: External content is logged directly without filtering for potential malicious instructions.

customerio-advanced-troubleshooting