The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The troubleshooting section instructs the user to execute sudo apt-get install, which grants root access to system package management and poses a high risk if the environment is not secured.
[EXTERNAL_DOWNLOADS]: The skill directs users to clone and download its components from an unverified GitHub repository (ColonyShopkeeper/r08-composiohq-awesome-claude-skills-seo), which is not associated with the recognized developer or trusted vendor lists.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process arbitrary web content through its auditing and SERP monitoring features, creating a surface for indirect instructions to influence agent behavior.
Ingestion points: Website URLs provided to commands like /content-audit, /technical-seo, and /page-speed-seo.
Boundary markers: The instructions do not define delimiters or markers to ensure the agent ignores malicious instructions embedded within crawled HTML or text.
Capability inventory: The skill possesses capabilities for network access, file system writing (via --file flags), and executing shell commands.
Sanitization: There is no mention of sanitizing or filtering external content before it is processed by the AI models or integrated into workflows.
[COMMAND_EXECUTION]: Provided shell script examples (seo-sprint.sh) interpolate data retrieved from external URLs (such as page titles or search result links) into command arguments, which could lead to command injection if the remote content contains malicious characters.
[METADATA_POISONING]: The skill metadata lists the author as ara.so and points to a repository named ColonyShopkeeper, which creates inconsistency with the stated vendor context of reason-machines.

composiohq-awesome-claude-skills-seo-content-marketing