The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes shell commands to perform its core functions.
Evidence: references/audit.md uses subprocess.run to call a Python audit script from the sibling seo-ops skill.
Evidence: references/links.md uses curl to verify link status and retrieve headers from audited sites.
[DYNAMIC_EXECUTION]: The skill generates and executes a script at runtime.
Evidence: references/audit.md uses a shell heredoc (`python3
<<'PY'`) to wrap the audit execution logic within a Python environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, creating an attack surface for indirect prompt injection.
Ingestion points: External site content and headers are retrieved from $SITE_URL and $LINK_URL via curl and the audit tool.
Boundary markers: The skill includes instructions in references/reports.md to avoid using backticks or dollar-brace sequences when generating HTML reports, which acts as a rudimentary boundary.
Capability inventory: The skill has the capability to execute shell commands, perform network requests to external domains, and write to the local filesystem via the EVIDENCE_DIR.
Sanitization: The instructions emphasize literal data handling and warn against inventing content not present in the source, though they do not include formal data sanitization logic.
[DATA_EXFILTRATION]: The skill performs network operations to external domains to retrieve site data.
Evidence: references/links.md uses curl to fetch content from dynamically resolved $LINK_URL values. While intended for auditing, this represents a network access capability that could be misused if combined with sensitive data access.

seo-workflows