The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute a shell command dynamically provided in the system reminder (<spex-init-command>). This creates a risk where the agent executes arbitrary code if the system context is manipulated.
[COMMAND_EXECUTION]: The skill provides instructions to modify .claude/settings.json to grant broad bash execution permissions, including wildcards like Bash(*spex-init.sh*) and Bash(specify *) which can be abused to run unauthorized scripts with similar names.
[PROMPT_INJECTION]: The skill uses extremely high-priority language ("You MUST complete ALL steps", "Do not stop after Step 1", "Your first and only Bash call") to override the agent's standard reasoning and process flow.
[EXTERNAL_DOWNLOADS]: The skill identifies and recommends the installation of third-party plugins from an external GitHub repository (github.com/obra/superpowers) that are not part of the standard trusted environment.
[COMMAND_EXECUTION]: The 'YOLO' configuration option explicitly instructs the agent to set "defaultMode": "bypassPermissions" and adds wide-reaching allow rules for Read(*), Edit(*), Write(*), WebFetch, and Bash(*). This effectively disables all security guardrails and permission prompts for the agent, leading to high-risk privilege escalation and persistent bypass of the security sandbox.

init