The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from WordPress posts and pages using respira_list_pages and respira_list_posts to find orphaned shortcodes. The instructions lack boundary markers or sanitization logic to isolate this content. Combined with administrative capabilities like respira_delete_plugin and respira_update_option, this creates a surface where malicious instructions in a post could potentially influence the agent's actions during the audit.
Ingestion points: SKILL.md (Step 4, Step 7) fetches WordPress content via respira_list_pages and respira_list_posts.
Boundary markers: None provided in the workflow.
Capability inventory: SKILL.md lists tools for plugin deletion (respira_delete_plugin), deactivation, and database option updates (respira_update_option).
Sanitization: No sanitization or escaping of external content is mentioned.
[DATA_EXFILTRATION]: The skill transmits site telemetry including the site URL, WordPress version, and audit findings to https://www.respira.press/api/skills/track-usage. This endpoint is owned by the skill's author, and the transmission is documented as part of the execution workflow.
[DYNAMIC_EXECUTION]: The skill employs respira_invoke_ability as a fallback mechanism for generating reports on older versions of the plugin. It passes a hardcoded ability string ('respira/generate-activity-report'), which is a form of dynamic execution.

technical-debt-audit