The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions require cloning source code from an unverified GitHub repository (https://github.com/kentjuno/ainovel-cli.git) that does not belong to the skill author or a trusted organization.
[REMOTE_CODE_EXECUTION]: Users are guided to compile and execute the external source code using go build and docker build, leading to the execution of unverified binaries on the host system or within a container.
[COMMAND_EXECUTION]: The skill provides numerous shell commands for tool execution, including batch processing scripts that run locally and mount host directories into a Docker environment, granting the tool access to the local file system.
[PROMPT_INJECTION]: The multi-agent architecture is susceptible to indirect prompt injection through its core inputs.
Ingestion points: User-provided novel prompts and the custom_rules.md file in the workspace directory.
Boundary markers: None identified in the provided instructions to separate instructions from generated content.
Capability inventory: The tool performs file system writes to the workspace/ directory, network requests to LLM providers (OpenRouter, Anthropic, Gemini, OpenAI), and execution of local CLI commands.
Sanitization: There are no documented mechanisms for sanitizing or escaping the user-provided prompts before they are processed by the agent orchestration loop.
[CREDENTIALS_UNSAFE]: The configuration examples involve the use of sensitive API keys (OPENROUTER_API_KEY, ANTHROPIC_API_KEY) which are passed via environment variables to the unverified external tool.

ainovel-cli-vietnamese