The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The scripts/inject-brand-context.cjs utility extracts brand voice and rules from docs/brand-guidelines.md and formats them into a 'BRAND CONTEXT' block for injection into the agent's prompt. This creates a vulnerability where malicious instructions in the guidelines file could manipulate agent behavior.
Ingestion points: docs/brand-guidelines.md is processed by scripts/inject-brand-context.cjs.
Boundary markers: Although the output is structured with headers, it lacks explicit safety delimiters to prevent the LLM from obeying instructions found within the data.
Capability inventory: The skill has the ability to write files to the assets/ directory and execute shell commands through the sync-brand-to-tokens.cjs script.
Sanitization: There is no sanitization or validation of the text extracted from the markdown file.
[COMMAND_EXECUTION]: The script scripts/sync-brand-to-tokens.cjs executes an external Node.js script located at .claude/skills/design-system/scripts/generate-tokens.cjs using execFileSync. Additionally, scripts/extract-colors.cjs provides instructions for running a Python script from the ai-multimodal skill directory. These patterns introduce dependencies on code execution from paths outside the skill's own directory.

brand