The Agent Skills Directory

[COMMAND_EXECUTION]: The skill constructs shell commands by interpolating user-provided input (<url>) directly into script arguments (e.g., claude-seo run render_page.py <url>). This pattern creates a risk of command injection if malicious strings are provided as the URL.
[DATA_EXFILTRATION]: The skill instructions involve checking for and using sensitive API credentials for Google, Moz, and Bing services. This interaction with authentication data and external APIs for data enrichment increases the risk of credential exposure or exfiltration if the execution environment is compromised.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests and processes large volumes of untrusted data from external websites.
Ingestion points: Crawls the homepage and up to 500 subpages, capturing raw and rendered HTML using render_page.py as specified in SKILL.md.
Boundary markers: The instructions lack explicit boundary markers or warnings to the agent to disregard instructions or scripts embedded within the retrieved website content.
Capability inventory: The agent can execute subprocesses via the claude-seo CLI, write files to the local file system (under {domain}-audit/), and make further network requests through its crawling functionality.
Sanitization: No evidence of sanitization or safety-filtering is provided for the data retrieved from external sources before it is analyzed by the agent or its subagents.

seo-audit