The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes instructions in references/codex-network.md designed to bypass system-level security controls. It specifically advises users to set approval_policy = "never" and enable network_access in the configuration, effectively disabling safety prompts and sandbox restrictions.
[PROMPT_INJECTION]: The file imagegenskill_audit_result_v1.json is a deceptive document masquerading as a pre-verified security audit result. It falsely claims the skill is "Production Ready" and has passed security gates, which is a self-referential attempt to influence security reviewers and automated scanners.
[EXTERNAL_DOWNLOADS]: The documentation in references/cli.md references the use of external Python packages openai and pillow for image generation, which contradicts the primary claim in SKILL.md that the skill avoids image models and external dependencies.

imagegenskill