The Agent Skills Directory

[PROMPT_INJECTION]: The skill architecture presents an indirect prompt injection surface as it processes external project configurations to generate deployment plans.
Ingestion points: Project files analyzed to detect stack, build commands, and environment variables as specified in the workflow in SKILL.md.
Boundary markers: No specific delimiters or safety instructions are defined to separate untrusted project data from the agent's instructions.
Capability inventory: The skill identifies build and deployment sequences and hands off results to a subsequent launch-readiness-auditor skill.
Sanitization: No data sanitization or validation mechanisms are described for the ingested project metadata.
[COMMAND_EXECUTION]: The SKILL.md file includes validation blocks that execute local Python scripts via shell commands from the $CODEX_HOME directory. These appear to be internal auditing tools used for skill verification.

deploy-orchestrator