The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains instructions that significantly modify the agent's operational mode toward high autonomy. It explicitly mandates that the agent should "Never Block on the Human" and follow a "Proceed, then present" pattern for work it considers "reversible." While the instructions provide a boundary for "irreversible actions," the shift in behavioral logic encourages the agent to bypass standard human-in-the-loop checkpoints, which may lead to unintended actions if the agent's judgment of reversibility differs from the user's.
[COMMAND_EXECUTION]: The instructions direct the agent to utilize external CLI workers and scripts, specifically mentioning the spawn-codex-worker script and codex exec command. These components are described as part of an "installed pstack skill" but are not provided in the analyzed file, indicating a dependency on an external execution environment with unverified scripts.

principle-never-block-on-the-human