The Agent Skills Directory

[SAFE]: The skill implements a comprehensive 'Runtime Trust' policy in references/runtime-trust.md that explicitly instructs the agent to treat all external data, repository files, and registry content as untrusted. It prohibits embedded instructions from altering the agent's governing rules or authority.
[SAFE]: The project scaffolding logic enforces industry-standard supply chain security by pinning dependencies exactly (e.g., react@19.2.8) and requiring the use of bun.lock for reproducible builds.
[EXTERNAL_DOWNLOADS]: The skill fetches package metadata from the official NPM registry (https://registry.npmjs.org) via the scripts/resolve-package-versions.ts utility. This script uses bounded JSON requests to verify package compatibility and security pins without executing remote code.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user requirements to scaffold new applications, presenting a surface for indirect prompt injection.
Ingestion points: User-supplied requirements and destination paths defined in the SKILL.md argument hint.
Boundary markers: Explicitly present in references/runtime-trust.md, which mandates that embedded instructions cannot alter governing rules or authority.
Capability inventory: The skill is authorized to execute bun for scaffolding, testing, and building, and oxlint/typescript for quality gates.
Sanitization: The agent is instructed to validate inputs before handler execution and uses encodeURIComponent in internal scripts to prevent URI-based injection when checking registry data.
[COMMAND_EXECUTION]: The skill executes standard development tools, including Bun, Vite, and Oxc, to perform project initialization and CI tasks. These executions are scoped to the project creation lifecycle and governed by the strict validation rules defined in the setup instructions.

tailrocks-tanstack-project-setup