The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface by processing data from untrusted sources to modify agent behavior.
Ingestion points: Reads demoReviewIssues from the CRM and scrapes external company websites using WebFetch (company.domainName.primaryLinkUrl).
Boundary markers: None identified; external content is directly used for rewriting prompts and knowledge items.
Capability inventory: The skill uses MCP tools to modify system state, specifically update_prompt to rewrite the demo's system instructions and add_to_bucket to update the knowledge base.
Sanitization: Instructions specify rewriting content based on scraped data but do not mandate sanitization or filtering to prevent malicious instructions from being incorporated into the demo agents.
[COMMAND_EXECUTION]: The skill uses curl to interact with the vendor's GraphQL CRM and the OpenBrand API. These are standard operations for the skill's intended purpose of demo management.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to openbrand.sh to extract brand colors. While this is an external service, it is used for the specific functional purpose of styling the demo widgets.

fix-demos