The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches software packages and configuration scripts from official vendor and trusted third-party repositories.
Downloads KASM release archives from the official kasm-static-content.s3.amazonaws.com storage bucket.
Retrieves the cloudflared binary from Cloudflare's official GitHub releases.
Configures and installs packages from the official Docker repository (download.docker.com).
[REMOTE_CODE_EXECUTION]: Provides instructions for downloading and executing installation and configuration scripts from trusted sources.
Facilitates the execution of the KASM install.sh script after downloading from official infrastructure.
Documents the installation and deployment of the cloudflared binary to establish a secure tunnel.
[COMMAND_EXECUTION]: Instructs the agent on performing administrative tasks using system-level commands.
Extensive use of sudo, docker, and psql for managing containers, system services, and database state.
Utilizes curl to interact with Cloudflare and KASM APIs for environment provisioning and management.
[PRIVILEGE_ESCALATION]: Documents configuration steps that grant or utilize elevated system permissions.
Includes instructions for modifying the sudoers configuration to grant passwordless sudo access to users within workspace containers.
Describes the modification of /etc/fstab for persistent swap file configuration.
Instructions for creating and managing systemd services and modifying file permissions across restricted directories like /opt/kasm/ and /mnt/.
[PERSISTENCE]: Documents methods for ensuring long-term operation of services and maintenance tasks.
Provides configuration and deployment steps for a systemd service to maintain the Cloudflare Tunnel connection.
Suggests crontab entries to schedule regular backup and synchronization tasks to external S3-compatible storage using rclone.
[INDIRECT_PROMPT_INJECTION]: The skill's workflow for determining user needs based on input creates a potential surface for indirect prompt injection.
Ingestion points: The agent parses user requests to determine which reference file to consult within SKILL.md.
Boundary markers: No explicit delimitation or warnings are provided to prevent embedded instructions in user queries from influencing agent behavior during file selection.
Capability inventory: The skill possesses significant capabilities for command execution, network requests, and filesystem modification across all reference files.
Sanitization: There is no documented validation or sanitization of user-provided task names before they are used to branch logic in the main skill file.

kasm