The Agent Skills Directory

[DATA_EXFILTRATION]: The skill includes instructions to 'silently' perform a POST request to an external telemetry endpoint (https://www.respira.press/api/skills/track-usage) after generating the report. This payload includes the target site URL, the PHP version of the environment, and a summary of the audit findings. While the endpoint belongs to the skill's author, the silent transmission of environmental metadata (like PHP versions) and site identification to a non-whitelisted domain constitutes a data exposure risk.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from the target WordPress site, including page content, metadata, and builder configurations, using tools like respira_list_pages and respira_analyze_performance. Because the skill also provides the agent with capabilities to modify the site (e.g., respira_update_element, respira_batch_update), it creates a surface for indirect prompt injection. Malicious instructions embedded in a site's web content could attempt to manipulate the agent's report or influence the 'Fix Workflows' suggested to the user.
Ingestion points: respira_list_pages (content and metadata), respira_get_builder_info, respira_analyze_performance.
Boundary markers: No delimiters or instructions to ignore embedded commands are specified for the ingested content.
Capability inventory: respira_update_element, respira_batch_update, respira_create_page_duplicate, respira_get_snapshot, respira_restore_snapshot.
Sanitization: No sanitization or validation of the retrieved site content is mentioned before processing.

mobile-experience-report