The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains no instructions designed to bypass safety filters, override system prompts, or extract internal agent instructions. The instructional language is professional and task-oriented.
[DATA_EXFILTRATION]: No evidence of data exfiltration was found. The skill accesses local project files related to game design (GDDs, architecture documents) to perform its task, which is the intended behavior. There are no network operations targeting external or non-whitelisted domains.
[CREDENTIALS_UNSAFE]: The skill does not contain hardcoded secrets, nor does it attempt to access sensitive system locations like SSH keys or environment variable files. It explicitly recommends using configuration files for gameplay values instead of hardcoding them.
[REMOTE_CODE_EXECUTION]: No remote code execution patterns were detected. The skill does not download external scripts or execute code from remote servers.
[COMMAND_EXECUTION]: The skill uses the 'Bash' and 'Task' tools to manage the build process and spawn a 'creative-director' agent for review. These activities are within the declared 'allowed-tools' and align with the skill's primary purpose of prototyping game builds.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user input and playtest feedback to generate technical reports. While these are external ingestion points, the skill operates within a controlled worktree environment and does not interpolate these inputs into high-privilege execution contexts. The risk is assessed as minimal and consistent with standard agentic functionality.

vertical-slice