The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external data such as product descriptions, competitor references, and marketing materials. This creates a surface for indirect prompt injection where instructions could be hidden in the product data.
Ingestion points: Product materials, packaging images, and reference documents specified in SKILL.md and examples/README.md.
Boundary markers: The skill explicitly implements an 'Evidence Ledger' (Step 1) to classify claims as Verified, Supported, Needs proof, or Disallowed. It also includes a 'Quality Gate' (Step 6) with hard blockers against fabricated data and instructions to 'never turn assumptions into facts'.
Capability inventory: The skill primarily generates text, storyboards, and image-generation prompts. No subprocess calls or file-system writes are present in the instruction set across ALL scripts.
Sanitization: The instructions require the agent to label assumptions and explicitly separate verified facts from unproven claims, mitigating the risk of instructions embedded in data being executed as factual directives.
[SAFE]: The repository includes a Python utility (tools/verify-skill.py) meant for developers to verify the skill's structure and check for accidental secret leaks before distribution. This script uses standard libraries and performs local file reads only, with no network or dangerous system operations detected. It specifically scans for common secret patterns (e.g., GitHub tokens, API keys) to ensure the developer's environment is safe.

ecommerce-visual-copywriting