The Agent Skills Directory

[COMMAND_EXECUTION]: The skill configuration explicitly enables high-risk execution modes for AI agents, such as --dangerously-skip-permissions and --yolo by default. This design choice allows agents to perform autonomous system-level actions without requiring user approval or manual verification.
[EXTERNAL_DOWNLOADS]: The primary execution and installation method is npx vibe-kanban, which dynamically retrieves and runs code from the npm registry. This introduces a dependency on the integrity of the remote package and registry security.
[DATA_EXFILTRATION]: Features a 'Remote Access' system where a local host machine can be paired with https://cloud.vibekanban.com using pairing codes. This facilitates remote interaction with local workspaces, which could potentially expose local development environments to a third-party cloud service.
[DATA_EXFILTRATION]: The tool automates the copying of sensitive files, such as .env files, from the main project into isolated git worktrees. The ability to disable automatic cleanup via the DISABLE_WORKTREE_CLEANUP environment variable increases the risk that these secrets may persist on the filesystem.
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection.
Ingestion points: Data is read from external repositories and GitHub/Azure DevOps pull requests (references/workspaces.md, references/core-features.md).
Boundary markers: No specific delimiters or safety instructions are mentioned for isolating or sanitizing untrusted data.
Capability inventory: Agents have access to the shell, file system, and various project management tools via the Model Context Protocol (references/integrations.md).
Sanitization: No documented process exists for sanitizing external inputs before they are processed by the AI agents.

vibekanban