The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The script scripts/env-loader.cjs is designed to load environment variables from the global agent configuration directory at $HOME/.claude/.env. Accessing platform-level configuration files may expose sensitive global secrets or API keys to the skill that are outside its intended scope.
[CREDENTIALS_UNSAFE]: In scripts/competitor-traffic.cjs, the SEMRUSH_API_KEY is transmitted as a query parameter in a GET request to the SEMrush API. Including credentials in URLs is a security risk as they can be exposed in server logs, proxy logs, and browser history.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection through the untrusted data it ingests from external websites.
Ingestion points: scripts/market-scraper.cjs scrapes content (HTML and Markdown) from arbitrary external URLs provided during the research workflow.
Boundary markers: The skill does not use boundary markers or specific delimiters to isolate scraped content when presenting it to the agent for analysis.
Capability inventory: The agent has the capability to execute local shell commands (Python and Node.js scripts) and invoke the gemini tool, which could be targeted by instructions hidden in scraped data.
Sanitization: The scraped data is parsed for specific fields but is not sanitized to remove potential malicious instructions or payloads.

marketing-research