The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to process external data from emails, chats, and meeting transcripts. To address the potential for indirect prompt injection, it includes specific instructions to treat all retrieved content as untrusted data and to ignore any directives found within those sources.
Ingestion points: Data enters the agent context through WorkIQ and board integrations (ADO/GitHub), as well as local transcript files.
Boundary markers: The skill requires an explicit notice to the user about data sensitivity and instructions to the agent to disregard embedded instructions.
Capability inventory: The skill possesses the ability to write to the local file system and interact with mail draft operations via MCP tools.
Sanitization: Instructions are provided to summarize findings and exclude raw sensitive information, credentials, or PII from reports.
Local Data Management: The workflow involves saving sensitive research findings to a local .working/ directory. The skill manages this risk by requiring the user to acknowledge a data sensitivity notice and ensuring that working directories are protected by ignore rules to prevent accidental exposure via version control.
Restricted Tool Operations: While the skill can interface with Outlook, it is strictly limited to creating drafts. It requires explicit user approval for this action and is explicitly prohibited from performing any automated send operations.

engagement-reporting