The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions and environment configuration facilitate the bypassing of critical safety guardrails and permission filters.
Evidence: The skill instructs that Claude-based worker agents should be launched with the --dangerously-skip-permissions flag, which disables security checkpoints for the sub-agents (found in SKILL.md).
Evidence: The RCS_TEAM_AUTO_ACCEPT_BYPASS environment variable is documented to auto-advance past security prompts by default, specifically automating the input 2 + Enter (found in SKILL.md).- [COMMAND_EXECUTION]: The skill establishes a mechanism for automated command execution across multiple persistent shell sessions.
Evidence: The implementation uses tmux send-keys to programmatically deliver triggers and instructions to worker panes (found in SKILL.md).
Evidence: The environment variable RCS_TEAM_WORKER_LAUNCH_ARGS allows for the injection of arbitrary command-line arguments into the initialization of worker agent processes (found in SKILL.md).
Evidence: The RCS_TEAM_WORKER_CLI and RCS_TEAM_WORKER_CLI_MAP variables determine the executable launched for each worker, allowing for the execution of different CLI tools based on environment configuration (found in SKILL.md).- [INDIRECT_PROMPT_INJECTION]: The skill processes external task data and context files to generate instructions for worker agents, creating an attack surface for instruction injection.
Ingestion points: Task descriptions provided via command-line arguments and markdown files stored in .rcs/context/ (found in SKILL.md).
Boundary markers: No explicit boundary markers or delimiters are defined to isolate user-provided task content from the worker agent's system instructions.
Capability inventory: Arbitrary shell command execution via tmux, filesystem write access for state management, and Git operations (found in SKILL.md).
Sanitization: The skill documentation mentions sanitizing team names from task text but does not describe validation or sanitization of the actual task content (found in SKILL.md).

team