The Agent Skills Directory

[PROMPT_INJECTION]: The skill explicitly instructs the agent to bypass standard safety protocols (specifically the 'Collaboration Protocol' referenced in the text) for file writes. It defines a 'bounded exception' where sub-agents are directed to write to specific paths without asking for user approval, stating 'Do not "fix" this by asking per subagent'.
[PRIVILEGE_ESCALATION]: By instructing the agent and its sub-agents to ignore mandatory confirmation requirements for file modifications, the skill attempts to escalate its operational capabilities beyond the platform's standard safety constraints.
[DYNAMIC_CONTEXT_INJECTION]: The skill utilizes dynamic context injection (!) to execute a bash script at load time. This execution happens automatically when the skill is accessed and uses path traversal (../../hooks/yaml-helper.sh) to execute a script located outside the skill's specific directory.
[COMMAND_EXECUTION]: The allowed-tools configuration for the Bash tool uses a path traversal pattern (*/.claude/skills/team-release/../../hooks/yaml-helper.sh) to target a script in a parent directory structure. This allows the skill to execute code from the broader environment rather than just its own packaged scripts.

team-release