The Agent Skills Directory

[DATA_EXFILTRATION]: The iMessage provider documentation describes a 'local' mode that reads the macOS Messages database (chat.db) directly from the file system. This allows the agent to process local conversations but grants access to sensitive personal communication history.
[EXTERNAL_DOWNLOADS]: The Terminal provider is documented to automatically download the tuichat binary from the author's GitHub repository (photon-hq/tuichat) upon initialization to provide a development and testing interface.
[PROMPT_INJECTION]: The skill facilitates the ingestion of messages from various external messaging platforms, which creates a surface for indirect prompt injection.
Ingestion points: Untrusted message content enters the agent context via the app.messages stream (documented across all topic files).
Boundary markers: No specific delimiters or 'ignore' instructions for embedded content are mentioned in the provided implementation samples.
Capability inventory: The agent can send messages, react, and reply across multiple platforms (space.send, message.react, message.reply).
Sanitization: No explicit sanitization or validation routines for incoming message text are described in the SDK patterns.

spectrum