The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides patterns for ingesting and processing inbound email content and attachments, which are untrusted external inputs.
Ingestion points: Untrusted data enters via the process method in mailbox classes which handle raw email bodies and attachments as documented in SKILL.md and reference.md.
Boundary markers: The instructions do not specify the use of delimiters or system-level warnings to distinguish between data and instructions when the agent processes email content.
Capability inventory: The skill's context allows file modification (Write, Edit) and shell command execution (Bash), creating a path for malicious instructions in emails to potentially influence the environment.
Sanitization: Content-focused sanitization methods like EmailReplyTrimmer and strip_tags are mentioned for display purposes, but do not provide protection against adversarial instructions targeted at the LLM.

action-mailbox