Elyro does not run a coding agent. Codex, Claude Code, or another host agent edits the project on macOS and uses Elyro only for Linux Workspace execution.
Install the embedded Skill:
elyro skill install codex
elyro skill install claude-code
elyro skill install allInspect it without installing:
elyro skill showIn a configured project without an installed Skill, a host agent that discovers elyro.yaml can inspect the Agent section of elyro --help, use elyro skill --help to find the inspection and installation paths, and run elyro skill show for the complete version-matched Skill. After the Skill activates, individual command help remains the source for exact syntax and available options.
The Skill applies only to a configured project: elyro.yaml must exist in the intended project directory or the nearest ancestor of the current working directory. This activation boundary applies to the Skill; it does not remove Elyro's zero-configuration CLI workflow for direct human use. The agent does not infer Elyro usage merely from macOS, a need for Linux validation, an installed Elyro binary, an existing Workspace, or a generic mention in agent instructions. It also does not scan unrelated descendants for configuration. If the precondition is met, the Skill directs the host agent to inspect elyro status --json, preview lifecycle changes with elyro up --dry-run --json, start a missing Workspace only when needed, and run Linux commands through elyro exec -- .... A recreate plan is surfaced to the user instead of being applied implicitly by the agent. Project runtime environment configuration is inherited automatically; a temporary per-command override can use elyro exec -- env KEY=value command. The Skill does not contain scripts, credentials, MCP configuration, or model settings.
## Linux validation
Use the installed `use-elyro-workspace` Skill only when this project's root contains `elyro.yaml`; when working in a nested directory, use the nearest ancestor containing that file. Otherwise, do not use the Skill or run Elyro commands. For a configured project, inspect `elyro status --json` first and preview changes with `elyro up --dry-run --json`. Start a missing Workspace with `elyro up --json`; ask before applying a `start` or `recreate` plan to an existing Workspace. Run commands with `elyro exec -- ...`; use `elyro exec -- bash -lc '...'` only for explicit shell syntax. Do not call Docker or SSH directly, do not create Elyro configuration from the Skill, and leave `elyro down` to the user.A Workspace is a convenient development environment, not a security sandbox. Elyro does not proxy model traffic, manage agent credentials, or replace the coding agent's permission model. Editor opening and destructive cleanup remain user decisions.