The Agent Skills Directory

[SAFE]: The skill operates as a tool-agnostic orchestrator. It does not call external rendering APIs, manage API keys, or access sensitive system credentials. Its primary function is the generation of Markdown artifacts (prompts and manifests) based on project-specific design briefs.
[COMMAND_EXECUTION]: The skill uses local Bash commands to run two utility scripts: scripts/log-critic-override.ts and scripts/manifest-sync.ts. These scripts perform standard logging and manifest indexing tasks within the project's own directory structure and do not perform network operations or access sensitive files outside the project scope.
[PROMPT_INJECTION]: The skill utilizes a structured two-agent workflow (Prompt Author and Critic) with nine explicit 'Critical Gates'. These gates enforce strict compliance with the source brief, explicitly forbidding the hallucination of brand marks, modification of copy, or unauthorized adjustments to aspect ratios.
[DATA_EXFILTRATION]: No network exfiltration patterns were detected. All network-related instructions in the skill body are documentation regarding target engines for the operator to use manually. The skill's scripts are local-first and do not contain code for remote data transmission.
[INDIRECT_PROMPT_INJECTION]: An attack surface exists as the skill ingests content from external brief artifacts (docs/forsvn/artifacts/marketing/design-briefs/[slug].md).
Ingestion points: Design brief artifacts specified by the operator.
Boundary markers: The skill uses clearly defined Markdown headers and 'verbatim' section markers in the generated prompts to isolate external content.
Capability inventory: Local script execution via Bun and file-system write access to the project's artifact folders.
Sanitization: The critic-agent.md performs binary validation of all output against the source brief and brand guidelines before any files are written to disk.

produce-asset