The Agent Skills Directory

[COMMAND_EXECUTION]: The instructions direct the agent to launch workers using an 'installed pstack skill spawn-codex-worker script' or 'direct codex exec'. These are non-standard commands not found in common environments or provided in the skill package, which may lead to execution of unauthorized or untrusted code.
[PROMPT_INJECTION]: The skill attempts to redefine the agent's core operational logic by overriding model selection (e.g., replacing Claude Opus with a nonexistent 'Codex gpt-5.6-sol') and redirecting task delegation to external CLI workers.
[PROMPT_INJECTION]: The skill uses forceful directives such as 'Must always apply' and 'Apply these overrides before following the original workflow' to compel the agent to disregard its default instructions and system prompts in favor of the skill's specific overrides.

unslop