The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The documentation in references/design-md-format.md recommends using npx design-md for linting and exporting design tokens. npx facilitates the download and execution of arbitrary code from the NPM registry. The package design-md is not a widely recognized standard tool and its attribution to 'Google Labs' cannot be verified, posing a risk of supply chain attack if the package name is typosquatted or the registry entry is compromised.
[INDIRECT_PROMPT_INJECTION]: The skill's workflow involves reading and parsing a DESIGN.md file from the project root to extract design tokens and rationale. This file acts as an untrusted external ingestion point.
Ingestion points: SKILL.md (Workflow Step 1) and references/design-md-format.md specify that the agent should parse DESIGN.md in the project root if it exists.
Boundary markers: The instructions lack delimiters or explicit warnings to the agent to ignore any natural language instructions or commands embedded within the DESIGN.md file.
Capability inventory: The skill is configured with a broad set of tools including Bash, Write, Edit, Grep, and WebFetch, which could be exploited if an attacker places malicious instructions in a DESIGN.md file.
Sanitization: There is no mention of sanitizing, validating, or filtering the content of the DESIGN.md file before the agent processes it.

frontend-design