The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill recommends a verification workflow that involves fetching JSON data from a local API and piping it directly to a Python interpreter for processing.
Evidence: curl -s http://127.0.0.1:8025/mailpit/api/v1/messages | python3 -m json.tool and curl -s http://127.0.0.1:8025/mailpit/api/v1/messages | python3 -c "...".
While targeting localhost (127.0.0.1) reduces the network attack surface, executing content fetched via network requests (even locally) is a dangerous practice as it bypasses standard safety boundaries.
[COMMAND_EXECUTION]: The skill provides numerous shell commands for environment management, including cleaning directories (rm -rf var/view_preprocessed/*), deploying static content, and running npm build scripts.
[PROMPT_INJECTION]: The skill creates an indirect prompt injection surface by encouraging the automated ingestion of email content from the Mailpit API without sanitization.
Ingestion points: Messages and HTML content are retrieved from http://127.0.0.1:8025/mailpit/api/v1/message/{MESSAGE_ID}/html.
Boundary markers: No delimiters or instructions to ignore embedded commands are used when retrieving or processing the email data.
Capability inventory: The skill suggests executing shell commands via ddev exec and running Python scripts based on the fetched data.
Sanitization: No sanitization or validation of the fetched HTML/JSON content is mentioned before it is processed or displayed.

email-theme-styling