The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill facilitates the installation of the official EffectCraft CLI and isolated Python runtimes. It uses a 'pinned' approach where binaries are verified against specific SHA-256 hashes (e.g., in runtime.lock.json and bridge-tools.json) before installation, preventing supply chain attacks.
[REMOTE_CODE_EXECUTION]: Execution is restricted to the specific, verified CLI binary. The skill uses a managed harness (effectcraft-harness) to launch commands through controlled shell/PowerShell scripts (launch.sh, launch.ps1), which ensures isolation and prevents unauthorized command injection.
[DYNAMIC_EXECUTION]: The skill processes user-defined JSON plans to drive graphics operations. It includes a dedicated validation step (commands.py check) that verifies JSON structure, command identity, and parameters against a fixed catalog before any execution occurs.
[INDIRECT_PROMPT_INJECTION]: The skill presents an ingestion surface for untrusted data (JSON plans). However, it mitigates risks by requiring explicit boundary markers (e.g., $ref for internal object references) and limiting the capability surface to specific graphics commands. It includes mandatory verification of saved projects and rendered outputs to ensure instructions were followed correctly.
[PRIVILEGE_ESCALATION]: The skill specifically avoids requesting administrative privileges. Documentation in desktop-install.md and managed-execution.md explicitly states that installation occurs in user-writable directories without modifying the system PATH or requiring 'sudo'.

effectcraft-cli-layers