The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructions require the agent to bypass standard safety and permission guardrails when spawning new sessions. It explicitly mandates the use of flags such as --allow-dangerously-skip-permissions for Claude-based harnesses and --yolo for Codex-based harnesses, effectively disabling safety protocols in the target environment.
[COMMAND_EXECUTION]: The skill performs shell command execution using CLI tools including herdr, claude, codex, and pi. These commands are used to manage workspaces, create tabs, and spawn new agent processes.
[DYNAMIC_EXECUTION]: The skill generates shell commands at runtime by concatenating user-selected models and harness configurations. It uses logic to derive CLI flags based on the chosen harness, such as -m <model> or identifiers reported by pi --list-models.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user data via the task argument and interpolates it into critical operational contexts without sanitization.
Ingestion points: The [task] argument provided by the user in SKILL.md.
Boundary markers: None; the task brief is passed directly to the new agent and used to derive operational identifiers.
Capability inventory: Execution of shell commands via herdr, claude, codex, and pi CLI tools.
Sanitization: None; the skill instructions direct the agent to "Derive a short tab label and unique agent name from the task," which creates a surface for command injection if the input contains shell metacharacters.

dispatch-new-session