The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions direct users to clone a repository from an unverified third-party GitHub account (kentjuno/ainovel-cli) rather than the stated author's infrastructure.
[REMOTE_CODE_EXECUTION]: Instructions guide the user to download source code and immediately compile it using 'go build' or build a Docker image, leading to the execution of unverified remote code on the host machine.
[COMMAND_EXECUTION]: The skill provides various shell commands for building, running, and diagnosing the tool, including complex Docker mounting operations and headless mode executions.
[PRIVILEGE_ESCALATION]: Troubleshooting steps suggest adding the user to the 'docker' group using 'sudo usermod -aG docker $USER', which provides the user with root-equivalent privileges on the host system.
[METADATA_POISONING]: The skill metadata identifies the author as 'ara.so', which differs from the vendor context 'reason-machines' provided in the instructions, creating potential confusion about the skill's origin.
[INDIRECT_PROMPT_INJECTION]: The skill processes external text files and user-supplied interventions to drive multi-agent behavior, creating a surface for indirect prompt injection.
Ingestion points: Reads from 'custom_rules.md', 'metadata.json', and processes real-time intervention strings and novel prompts.
Boundary markers: The instructions do not specify the use of delimiters or 'ignore' instructions to separate user-provided rules from the agent's core instructions.
Capability inventory: The skill has the capability to write various markdown and JSON files to the 'workspace/' directory and perform network requests to multiple LLM providers.
Sanitization: There is no mention of input validation or sanitization for the content processed from 'custom_rules.md' or intervention inputs.

ainovel-cli-vietnamese