The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides installation instructions for the Mailpit tool using a 'curl-to-pipe' shell execution pattern. This bypasses standard package verification and executes remote content directly.
Evidence: sudo sh < <(curl -sL https://raw.githubusercontent.com/axllent/mailpit/develop/install.sh) in SKILL.md.
[PRIVILEGE_ESCALATION]: The recommended remote installation script is executed with sudo, providing root access to the host system for any code fetched from the network.
Evidence: Use of sudo sh and sudo INSTALL_PATH=/usr/bin in SKILL.md.
[COMMAND_EXECUTION]: The skill documents the use of various CLI tools for installation and service management.
Evidence: brew install mailpit, go install github.com/mailhog/MailHog@latest, docker run -d ... mailhog/mailhog.
[EXTERNAL_DOWNLOADS]: The skill downloads binaries and scripts from GitHub and references third-party services like mail-tester.com and dmarcian.com for deliverability testing.
[INDIRECT_PROMPT_INJECTION]: The evaluation files provided with the skill (evals/) explicitly detail numerous vulnerability surfaces for indirect prompt injection.
Ingestion points: The skill facilitates retrieving email content (headers, subjects, and bodies) via the Mailpit REST API (/api/v1/search and /api/v1/message/).
Boundary markers: Several code examples lack sufficient delimiters when interpolating untrusted email data into test assertions.
Capability inventory: The skill has the capability to perform network requests, execute shell commands, and interact with local services.
Sanitization: Documentation within the evaluations identifies critical gaps where data from external mail is trusted without verification (e.g., trust-chain issues with DKIM headers as discussed in evals/auditor-checklist...).

mailpit-testing