The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides a suite of CLI tools (bunli dev, bunli build, bunli test, bunli release) to manage the application lifecycle. Command handlers have access to the shell via Bun.$ to perform legitimate development tasks.
[EXTERNAL_DOWNLOADS]: The project initialization tool (create-bunli) allows developers to fetch project templates from GitHub using the user/repo syntax. This is documented behavior for scaffolding new projects.
[PROMPT_INJECTION]: The framework handles external data which presents a surface for indirect prompt injection.
Ingestion points: CLI arguments, interactive prompts in references/utils/prompts.md, and remote templates in references/cli/scaffold.md.
Boundary markers: The framework integrates Zod and StandardSchema for defining explicit input boundaries.
Capability inventory: Skill handlers can execute shell commands, read/write files, and perform network operations during release processes.
Sanitization: Input is sanitized and validated through schema-driven APIs (option(), prompt(), Form).
[SAFE]: The skill includes functionality to detect if it is running in an AI-assisted environment (@bunli/plugin-ai-detect) for context-aware behavior, which is a standard feature for modern developer tools.

bunli