The Agent Skills Directory

[COMMAND_EXECUTION]: The skill initiates local script execution for batch image generation. The process uses runtime/codex_exec_image_batch.py and is designed with security constraints, such as disabling interactive stdin and operating within defined workspace limits as described in the Codex execution reference.
[PROMPT_INJECTION]: The skill manages user-provided creative briefs which are interpolated into routing prompts for sub-skills.
Ingestion points: User-supplied business context via chat or the intake.context field in SKILL.md.
Boundary markers: Internal routing uses specific command prefixes like $ads-explorer or $shot-explorer to scope subsequent agent actions.
Capability inventory: The system can execute local Python scripts for image generation and write configuration manifests to the workspace.
Sanitization: A "Brief Completeness Gate" verifies the presence of mandatory context (product, goal, audience, market) before rendering the Explore surface, preventing the processing of vague or malformed inputs.
[SAFE]: The skill documentation explicitly prohibits the use of arbitrary external image URLs in widgets to comply with restrictive resource policies, demonstrating a proactive approach to preventing common web-based vulnerabilities in the agent UI. No evidence of data exfiltration, credential exposure, or malicious obfuscation was found during the analysis.

creative-production-explore