The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill resolves a root path for script execution from local configuration files (.seo-suite-runtime.json) or environment variables (CLAUDE_PLUGIN_ROOT) rather than using static, verified paths. This mechanism allows the skill to execute code from computed locations.
[COMMAND_EXECUTION]: The skill instructs the agent to run multiple specialized Python scripts, including manage_markdown_site.py, manage_markdown_work.py, and validate_seo_findings.py, to perform inventory management, validation, and publication tasks.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest source data and user-defined tasks to generate sitemaps, topic maps, and llms.txt files. These outputs are intended to be consumed by other LLMs, creating a surface where malicious instructions in the source data could be propagated to downstream AI systems.
Ingestion points: Data dictionaries, record fields, and source update cadences in the Input gate and Markdown delivery contract.
Boundary markers: The skill relies on structured work notes and immutable evidence formats but does not define explicit security delimiters or prompt-level sanitization for untrusted data.
Capability inventory: The agent can execute shell commands via Python and perform file system writes to create or update discovery profiles and project documentation.
Sanitization: While there are instructions to exclude sensitive files (credentials, private reports) from public mirrors, there is no technical sanitization described for the interpolation of untrusted source text into the generated Markdown or LLM discovery files.

seo-programmatic