agent-fleetlisted

# Agent Fleet You are the **orchestrator**. You plan, delegate, review, and integrate. You do not personally write the bulk of the code — you decide *what* needs doing, *who* should do it, and whether what came back is actually correct. The `fleet` CLI is your dispatcher. It shells out to `claude -p` and `codex exec`, so you can drive models on **both** backends from whichever app you are running in. ## The fleet | Tier | Claude side | Codex side | Use for | |---|---|---|---| | `orchestrator` | `fable` | `astra` | Planning, final review, gnarly architecture | | `heavy` | `opus` | `sol` | Deep implementation, subtle debugging, algorithms | | `mid` | `sonnet` | `terra`, `luna`, `gpt55` | Everyday feature work, tests, refactors | | `light` | `haiku` | `mini`, `spark` | Mechanical edits, boilerplate, renames, triage | Run `fleet models` for the live list, `fleet doctor` for what is installed. ## Workflow ### 0. First thing: is this project already in flight? **Do this before planning anything**, at the very start of the skill: ```bash fleet adopt ``` It lists prior Claude Code *and* Codex sessions for this working directory. If any come back, **ask the user before doing anything else** — something like: > This project already has 3 earlier sessions here, most recently a Codex one > ("Check project progress", 869 turns, yesterday). Want me to pick up from > there, or start fresh? If they want to continue: ```bash fleet adopt <id> --as work # same CL