The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on multiple local Python scripts (such as backlinks_auth.py, moz_api.py, and verify_backlinks.py) to perform backlink auditing tasks. It also includes a shell script (./extensions/dataforseo/install.sh) meant to install an optional premium extension, which represents a local command execution pattern for tool management.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8) because it ingests and processes untrusted data from the open web, including anchor text, page content, and referring domain metrics.
Ingestion points: External data is ingested through scripts/verify_backlinks.py (which crawls targets), scripts/commoncrawl_graph.py (which processes web graph data), and various SEO APIs.
Boundary markers: Absent. The instructions do not specify the use of delimiters or 'ignore embedded instructions' warnings when the agent processes the retrieved SEO data.
Capability inventory: The skill has the capability to execute shell commands, run Python scripts, and perform network requests to external domains.
Sanitization: While the 'Pre-Delivery Review' checklist includes some structural validation (such as checking JSON-LD types and JS-rendering detection), it lacks specific security sanitization to filter or escape potential malicious instructions embedded in the external content.

seo-backlinks