The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external emails to analyze writing styles and relationship dynamics. This data is interpolated into prompts to generate email drafts, creating a potential vector for indirect prompt injection if an incoming email contains adversarial instructions designed to influence the agent's output or tool usage.\n
Ingestion points: External email content (bodies and metadata) retrieved via the mail-connector tool for the last 10 relevant messages.\n
Boundary markers: The skill documentation prescribes the use of summarized, verifiable signals and paraphrased evidence instead of raw content, which mitigates direct injection but does not eliminate the surface. Explicit delimiters for prompt construction are not defined in the provided files.\n
Capability inventory: The skill utilizes the mailc CLI tool to access and process email context.\n
Sanitization: Strict data minimization rules are enforced, including a prohibition on storing raw email bodies, long quotes, or unnecessary personal details in the local kontaktprofile.json store.

privat-mail-writer