The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external URLs during the research phase, which could contain malicious instructions designed to influence the agent.
Ingestion points: references/research.md specifies capturing public page HTML and screenshots via the firecrawl tool for competitor research.
Boundary markers: The skill relies on a system of "independent reviewer receipts" and "decision owner" approvals defined in SKILL.md to gate transitions between phases, acting as a manual check against automated obedience to injected instructions.
Capability inventory: The skill utilizes agent-browser for rendered proof, firecrawl for scraping, and various shell commands for building, linting, and testing as described in SKILL.md and references/verification.md.
Sanitization: references/design-and-data.md describes mapping raw service errors through approved product copy, and templates/shared-scenario-data-contract.md includes a requirement for a "Secret scan command" to identify sensitive data leaks.
[EXTERNAL_DOWNLOADS]: The skill process involves sourcing components from external registries and repositories. references/component-sourcing.md describes using registry commands to fetch files and installing packages as part of the sourcing workflow.
[COMMAND_EXECUTION]: The skill requires the agent to execute various system commands for environment setup, building, linting, and live verification, such as agent-browser, design-qa tools, and project-specific startup commands defined in SKILL.md and references/verification.md.

build-design-library