The Agent Skills Directory

[DATA_EXPOSURE]: The skill provides strong protections against data exposure. It includes a pre-configured .gitignore that excludes sensitive files like .p8 keys, .env files, keystores, and credentials. Additionally, it implements a no-secrets Lefthook job and specific agent deny rules in .claude/settings.json to prevent the AI agent from reading or committing signing material.
[COMMAND_EXECUTION]: The skill uses spawnSync to execute standard development tools like git, npm, and npx. Security is maintained by using the --no flag with npx, ensuring that only locally installed versions of tools like Prettier or ESLint are run, preventing the automatic download and execution of arbitrary remote packages.
[REMOTE_CODE_EXECUTION]: All package dependencies are managed through a strict dependency-policy.json and package-lock.json. The skill requires manual approval of install scripts one package at a time, protecting against malicious supply-chain attacks that use post-install scripts.
[PROMPT_INJECTION]: The skill instructions and agent contracts (AGENTS.md) contain natural instructional language intended to maintain project quality. There are no attempts to bypass safety filters or override system instructions for malicious purposes.
[SAFE]: The skill uses a 'guardrail' mechanism (check-quality-gates.ts) to ensure that security and quality configurations (like ESLint rules or agent permissions) are not silently weakened by the agent during a session.

expo-project-bootstrap