The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The scaffolding utility create-bunli (referenced in references/cli/scaffold.md) allows users to initialize projects using remote templates from arbitrary sources via the --template user/repo or github:user/repo syntax. This represents a vector for downloading and executing unverified code from third-party sources.
[COMMAND_EXECUTION]: The framework's command handler context (documented in references/core/commands.md) provides direct access to the shell via Bun.$. This high-privilege interface allows for the execution of arbitrary system commands, which could be exploited if user-supplied flags or arguments are not correctly sanitized before being passed to the shell.
[DATA_EXFILTRATION]: The bunli release command is designed to interact with external services, including npm and GitHub, for publishing packages and creating releases, involving the transmission of project metadata and potentially credentials.
[PROMPT_INJECTION]: The @bunli/plugin-ai-detect plugin (aiAgentPlugin) is specifically designed to detect the presence of AI coding assistants by inspecting environment variables (e.g., isAIAgent). This fingerprinting capability could be used to conditionally alter behavior when an AI is detected.
[PROMPT_INJECTION]: The skill processes external data that could serve as a vector for indirect injection attacks.
Ingestion points: CLI flags, positional arguments, and external configuration files such as bunli.config.ts (as described in references/cli/commands.md).
Boundary markers: The framework utilizes Zod schemas for type validation and coercion, though these do not act as security boundaries against instruction injection.
Capability inventory: Command handlers have extensive capabilities, including shell access (Bun.$), environment variable access (env), and file system interaction (cwd).
Sanitization: While the skill uses Zod for type-safe input parsing, it does not provide built-in mechanisms for escaping or sanitizing string data before it is used in high-privilege operations like shell execution.

bunli