The Agent Skills Directory

[SAFE]: The skill implements a local state management system using .claude/jupi.local.json to store workspace slugs and a contact map. This approach is consistent with standard practices for CLI-based tools to persist configuration and minimize repetitive user input.
[SAFE]: Interaction with external services is restricted to the Jupi platform using dedicated tools (create-decision-tool, add-decision-options-tool). The skill includes privacy-conscious defaults, such as setting allowWorkspaceContributions to false to limit data visibility.
[PROMPT_INJECTION]: The skill processes user-supplied information (decision context and alternatives) and formats it as HTML for submission to an external platform. This creates a surface for indirect prompt injection where untrusted data could include malicious HTML tags.
Ingestion points: User-provided decision titles, descriptions, and option details (SKILL.md).
Boundary markers: Absent; user input is interpolated into HTML strings for the description and options fields.
Capability inventory: External tool calls create-decision-tool and add-decision-options-tool (SKILL.md).
Sanitization: Absent; the skill relies on the agent to correctly format user prose into HTML for the target platform.

submit-decision