The Agent Skills Directory

[PROMPT_INJECTION] (LOW): Identified a surface for indirect prompt injection via the combination of data ingestion and high-privilege tool access.
Ingestion points: The skill explicitly instructs the agent to use WebFetch to retrieve content from external documentation sites and GitHub repositories (see references/documentation-map.md).
Boundary markers: No specific boundary markers or instructions to isolate external data from system instructions are provided in the skill body.
Capability inventory: The skill allows access to high-impact tools including Bash, Write, and Edit as defined in SKILL.md.
Sanitization: There is no evidence of sanitization or content validation for the data fetched from external URLs.
[DATA_EXFILTRATION] (LOW): The skill documentation suggests network operations to domains outside the predefined safe whitelist.
Evidence: The skill facilitates WebFetch calls to tanstack.com, zod.dev, and trpc.io. While these are relevant documentation sources, they are not on the core exfiltration whitelist.
[COMMAND_EXECUTION] (SAFE): The skill documentation suggests the use of standard development and deployment CLI tools.
Evidence: Commands such as npm install, npx wrangler pages deploy, and docker build are documented for their intended development purposes and do not show malicious intent.
[CREDENTIALS_UNSAFE] (SAFE): The skill correctly demonstrates secure secret management.
Evidence: Code examples in references/authentication.md and references/environment-variables.md emphasize the use of environment variables for secrets and provide guidance on .gitignore usage to prevent credential leaks.

tanstack-start