The Agent Skills Directory

[PROMPT_INJECTION]: The skill researches information via external search tools (Tavily, Exa), creating a surface for indirect prompt injection.
Ingestion points: External data is retrieved via belt app run commands and processed within the agent context (SKILL.md).
Boundary markers: The instructions do not provide explicit delimiters or guidance to ignore malicious instructions within the research results.
Capability inventory: The agent is granted Bash(belt *) permissions, allowing network interaction and tool execution.
Sanitization: There is no documented validation or filtering of the content retrieved from search queries.
[EXTERNAL_DOWNLOADS]: The skill references the 'belt' CLI and related skills from official repositories. These are standard dependencies for the skill's research-based features.
[COMMAND_EXECUTION]: The skill uses belt commands for authentication and running search modules, which is consistent with its stated purpose.

press-release-writing