The Agent Skills Directory

Local Command Execution: The skill instructs the agent to execute specific Node.js scripts (e.g., /workspace/tools/brand-lint.mjs, /workspace/tools/artifact-contract.mjs) to validate and lint generated assets. These commands use hardcoded paths to internal tools, which is a standard pattern for quality assurance within the agent's workflow.
File System Access: The skill defines a structured environment within /workspace/projects/ and references files in /workspace/brand/ and /opt/design-agent/. This is necessary for managing project artifacts and accessing design tokens or logos required for production.
Indirect Prompt Injection Surface: The skill processes user-provided requests ("the ask") and incorporates them into briefs and plans, creating a potential surface for indirect prompt injection.
Ingestion points: User-provided production requests and file attachments processed in SKILL.md.
Boundary markers: None explicitly defined for segmenting user input from system instructions.
Capability inventory: The skill utilizes node for script execution and has read/write access to the /workspace directory.
Sanitization: No explicit sanitization or filtering of input content is mentioned, though the skill explicitly notes that the execution sandbox has no network access, which significantly limits the risk of data exfiltration.

media-production