The Agent Skills Directory

[COMMAND_EXECUTION]: The skill dynamically constructs and executes shell commands using parameters provided by the caller, specifically for starting agents and managing worktrees. Evidence includes commands like herdr agent start <name> --kind <kind> --pane <pane_id> -- <agent flags> where <agent> and <agent flags> are variable inputs.
[INDIRECT_PROMPT_INJECTION]: The skill ingests a starting prompt from an upstream calling skill and relays it to a downstream agent session, creating a path for potential instruction injection.
Ingestion points: The starting prompt parameter in SKILL.md provided by /handover or /init-effort calls.
Boundary markers: The skill uses single-quote wrapping for the prompt string at the shell level but does not implement logical delimiters for the prompt content itself to differentiate between system instructions and untrusted data.
Capability inventory: The skill possesses capabilities to write to agent panes (herdr agent prompt), start new processes (herdr agent start), and manipulate file system workspaces via git worktrees.
Sanitization: The instructions include a specific shell-level sanitization step: write each ' inside it as '\'' to prevent shell-level command injection, showing awareness of basic injection risks.
[DYNAMIC_EXECUTION]: The skill implements dynamic execution patterns by taking command aliases (like <agent> for claude or codex) and flags from parameters to spawn new sessions, which is the core functionality of the handover process.

handover-to-herdr