The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of backend modules that ingest and process untrusted data from web forms, search queries, and external webhooks, which could theoretically be used for injection attacks against the agent or the application.
Ingestion points: Data enters the system through the act_submit handler in templates/php/lib/app.php, the act_hook handler in examples/r3-webhook-queue/lib/app.php, and query parameters in the search recipe (R7).
Capability inventory: The skill uses file system operations (file_put_contents), database interactions (db()), and outbound network requests (http_request, tg()) to handle the processed data.
Boundary markers: Instructions explicitly require the use of require_key() for cron and admin endpoints, as well as require_same_origin() for state-changing requests to mitigate unauthorized access.
Sanitization: The provided templates and recipes implement sanitization and validation using standard functions such as filter_var, trim, mb_strlen, strip_tags, and html_entity_decode before data is stored or echoed.

ucoz-server-scripts-skill