The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill executes a remote installation script by piping curl output directly into bash (e.g., curl -fsSL https://raw.githubusercontent.com/coleam00/Archon/main/scripts/install.sh | bash). This occurs in local macOS contexts and on remote Linux targets.
[COMMAND_EXECUTION]: Extensively uses SSH to execute arbitrary commands on remote VPS systems for environment discovery and smoke testing.
[COMMAND_EXECUTION]: Utilizes sudo on remote systems during the cleanup phase to remove files from privileged directories such as /usr/local/bin/.
[EXTERNAL_DOWNLOADS]: Downloads executable scripts and release artifacts from GitHub repositories. While these resources belong to the software author, the execution of unverified remote code is a significant security risk.
[PROMPT_INJECTION]: The skill has a surface for indirect prompt injection via manipulated GitHub release tags. Evidence Chain: 1. Ingestion points: Release version metadata fetched via gh release list. 2. Boundary markers: Explicit user confirmation required before starting the install/test phase. 3. Capability inventory: Remote code execution, SSH command execution, file system modification, and privilege escalation via sudo. 4. Sanitization: Absent for version strings interpolated into shell commands.

test-release