The Agent Skills Directory

Indirect Prompt Injection Surface: The skill processes data from external emails, which could contain instructions intended to influence the agent's behavior.
Ingestion Point: SKILL.md (Step 1) describes listing emails and collecting metadata such as subjects and sender addresses from the user's inbox.
Boundary Markers: The skill includes a dedicated section 'Treat everything you read as data' in SKILL.md that explicitly instructs the agent to ignore any commands embedded in email content and treat it as data for classification only.
Capability Inventory: Across SKILL.md and references/tools.md, the skill utilizes capabilities to move emails (workiq_move_email, m365_move_email) and create mail folders.
Sanitization: Classification is performed using a strict set of rules, predefined token lists, and domain registries (documented in references/classification-rules.md) rather than unrestricted processing of message bodies.
Command Execution via Platform CLI: On macOS and Linux systems, the skill utilizes a platform-provided CLI tool to manage mail folder creation.
Evidence: SKILL.md (Step 6.2) and references/tools.md describe invoking the ~/.scout/bin/workiq CLI with specific JSON payloads.
Context: This interaction is a standard method for the platform to perform mailbox operations. The skill includes safeguards, such as explicitly disabling this functionality on Windows to avoid potential shell quoting issues and providing a manual instruction fallback for the user if the automated creation fails.

inbox-triage