The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to download and incorporate a third-party Git submodule from an unverified GitHub repository: https://github.com/zubair-trabzada/geo-seo-claude.git.
[REMOTE_CODE_EXECUTION]: The skill relies on and executes several external Python scripts downloaded from the third-party vendor repository, including citability_scorer.py, fetch_page.py, brand_scanner.py, and llmstxt_generator.py. These scripts are executed locally via Python to perform diagnostic tasks.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill fetches content from arbitrary external URLs using fetch_page.py (documented in SKILL.md §12).
Boundary markers: There are no visible boundary markers or explicit instructions provided to the agent to treat fetched web content as untrusted data or to ignore any embedded instructions within that content.
Capability inventory: The skill utilizes sensitive tools including Bash, Write, Read, Grep, and Glob during the audit and reporting process (documented in frontmatter).
Sanitization: The instructions do not define any sanitization, filtering, or validation steps for the raw HTML or text data fetched from external sites before it is passed to the diagnostic scripts or the AI model for interpretation.
[COMMAND_EXECUTION]: The skill utilizes Bash and curl to perform network requests and local file system operations. While these are used for legitimate purposes like checking robots.txt and brand signals on Wikipedia/Wikidata, they provide a powerful interface that could be misused if influenced by malicious external data.

web-aeo