The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructions direct the agent to read from ~/.uplika//voice.md. Accessing files in the user's home directory outside of the immediate project scope is a potential data exposure risk, although this specific path is used for local configuration related to the authoring vendor.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user input to generate output drafts, creating a potential surface for indirect prompt injection. * Ingestion points: User input requested for 'Who', 'The trigger', and 'What the person wants' fields in SKILL.md. * Boundary markers: Absent; the skill does not specify delimiters for user-provided context. * Capability inventory: Text drafting and potential hand-off to 'inbox' or 'automations' tools for delivery. * Sanitization: Absent; no explicit instructions for filtering or escaping input data are provided.

ig-dm