The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill documentation and configuration files explicitly promote and enable high-privilege execution modes for AI agents.
Agent profiles in references/configuration.md include flags such as dangerously_skip_permissions: true, yolo: true, and sandbox: "danger-full-access".
These configurations are designed to bypass safety confirmations and prompts, granting the agents autonomous system access.
[COMMAND_EXECUTION]: The skill is designed to execute arbitrary shell commands and scripts on the host system.
references/core-features.md describes the use of 'Setup Scripts', 'Dev Server Scripts', and 'Cleanup Scripts' (e.g., npm install, cargo build) that run within git worktrees.
An integrated PTY-backed terminal with full shell support is provided within the workspace interface as noted in references/workspaces.md.
The MCP server integration allows external clients to execute workspace and issue management commands, including script updates.
[EXTERNAL_DOWNLOADS]: The skill instructions direct the user to download and execute code from external sources.
SKILL.md and references/getting-started.md provide commands to install the core application via npx vibe-kanban and vibe-kanban-web-companion via npm.
It encourages integration with third-party CLI tools such as the GitHub CLI (gh) and Azure CLI (az).
[DYNAMIC_EXECUTION]: The skill allows for the dynamic modification and execution of agent commands.
references/configuration.md details how users can provide base_command_override and additional_params to change how the underlying agent CLI is invoked at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data that could be used to influence the behavior of the autonomous agents.
Ingestion points: Task descriptions, code review comments, and arbitrary file attachments (logs, specs, archives) processed in SKILL.md and references/core-features.md.
Boundary markers: The documentation does not specify the use of delimiters or warnings to prevent agents from following instructions embedded within these data sources.
Capability inventory: The agents have broad capabilities including file system modification, git operations (merge, rebase, push), and execution of arbitrary scripts as described in references/core-features.md and references/workspaces.md.
Sanitization: No evidence is provided regarding the sanitization or validation of input data before it is passed to the agent's context.

vibekanban