The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill invokes npx vibe-kanban, which dynamically retrieves the application from the npm registry. It also references remote endpoints at https://api.vibekanban.com and https://vibekanban.online for remote connectivity and status updates.
[COMMAND_EXECUTION]: Shell scripts are used to manage the git environment and local configurations. The scripts/mcp-setup.sh script modifies the user's home directory configuration files for Claude (~/.claude/claude_desktop_config.json) and Codex (~/.codex/config.toml) to enable the MCP server integration. scripts/cleanup.sh uses git worktree and git branch commands to maintain workspace hygiene.
[REMOTE_CODE_EXECUTION]: The MCP integration is configured to execute npx vibe-kanban --mcp, resulting in the execution of code fetched at runtime from the npm registry.
[PROMPT_INJECTION]: The skill presents an indirect prompt injection surface (Category 8) due to its role in orchestrating agent actions based on task descriptions.
Ingestion points: Task descriptions are ingested via the Kanban board UI (SKILL.md) and the vk_create_card MCP tool (mcp-api.md).
Boundary markers: Absent; there are no explicit delimiters or safety instructions in the configuration scripts to isolate or ignore instructions within task metadata.
Capability inventory: The skill spawns CLI-based AI agents (claude, gemini, etc.), manages the local file system via git worktree, and executes system commands via npx.
Sanitization: No input sanitization is performed within the provided shell scripts, delegating safety to the underlying agents' internal guardrails.

vibe-kanban