The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes the arkcli binary to manage local files and directories (e.g., arkcli +connect, arkcli +connect uninstall). These operations are scoped to specific AI agent skill directories on the local machine.
[INDIRECT_PROMPT_INJECTION]: The skill serves as a management interface that writes files to the filesystem based on an internal catalog. It includes logic to resolve routing conflicts by removing specific third-party skills (byted-ark-seedance-skill, byted-ark-seedream-skill), which is documented as a functional requirement for the vendor's ecosystem. Mandatory Evidence Chain: 1. Ingestion points: Local filesystem structure and internal CLI catalog (SKILL.md). 2. Boundary markers: Absent (standard filesystem operations). 3. Capability inventory: Directory creation, file writing, and targeted deletion (SKILL.md). 4. Sanitization: Absent.
[SAFE]: The documentation explicitly states that the operations are purely local and do not require network access or authentication. No evidence of obfuscation, credential harvesting, or remote code execution from untrusted sources was detected.

arkcli-connect