The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides explicit instructions and code examples for agents to execute shell commands, including git worktree add for managing shared workspaces and various hermes kanban CLI verbs for task management.
[INDIRECT_PROMPT_INJECTION]: The core workflow relies on agents reading summary and metadata fields from previous workers via the kanban_show tool and the kanban_comment history. A compromised or malicious worker could inject instructions into these fields to influence the behavior of downstream agents.
Ingestion points: The runs array returned by kanban_show (containing summary, metadata, and outcome) and the kanban_comment thread.
Boundary markers: The skill does not define specific delimiters or instructions to ignore embedded commands within handoff data.
Capability inventory: Workers have permissions to write to shared persistent directories (dir:<path>), modify git worktrees, and create new tasks via kanban_create.
Sanitization: No sanitization or validation is specified for the free-text summary or structured metadata fields before they are presented to subsequent workers.

kanban-worker