The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill manages multi-tenant platforms where tenant-controlled data (e.g., project names, branch names, and user metadata) is ingested and used as parameters in management operations.
Ingestion points: Environment variables in the scripts/ directory (such as NEON_PROJECT_NAME, NEON_SNAPSHOT_NAME, USER_EMAIL) and parameters passed to agent tools defined in SKILL.md serve as potential entry points for untrusted data from platform tenants.
Boundary markers: The skill provides explicit security patterns for managing these boundaries, notably through the @neon/tools inject configuration which allows developers to hard-code projectId values and omit them from the agent's visible schema, effectively preventing agents from accessing other tenants' resources.
Capability inventory: The skill possesses significant control-plane capabilities via the @neon/sdk, including the ability to create, delete, and modify projects, branches, and database users across a fleet.
Sanitization: While the sample scripts rely on the Neon Management API's internal validation for input sanitization, the skill documentation emphasizes the use of project-scoped API keys to strictly limit the blast radius of any potential injection.
[CREDENTIALS_UNSAFE]: The skill correctly manages credentials by using environment variables (NEON_API_KEY) and explicitly documenting the use of project-scoped keys for agent runtime to ensure least-privilege access. It avoids hardcoding secrets, providing .env.example patterns for secure local development instead.
[EXTERNAL_DOWNLOADS]: All external references and dependencies target trusted vendor resources, specifically official Neon domains (neon.com, neon.tech) and the official @neon/sdk and @neon/tools packages. These references are documented neutrally and follow standard platform integration patterns.

neon-postgres-agent-platforms