# Model and Effort Each agent's model and effort are visible where you already look: inline in its pane header, and as a hover on its row in the agent list. The layout and the look of the rows don't otherwise change. ![Model and Effort](preview.png) ## What you see - Pane header, left to right: program icon, name, then the model text such as `Opus 5.5 · medium` in `t.muted`, then the folder in `t.dim`. The model text doesn't shrink; the folder truncates first. - Agent list: rows look the same. Hovering a row shows a tooltip with the same text. This also applies to the narrow, collapsed rows. - With no effort reported, only the model shows. With no model reported, nothing is added. ## How it behaves - Agent status (working, blocked, done, idle) doesn't affect the text. - Values come from three labels on the agent's terminal: `herdr.model`, `herdr.effort` and `herdr.model_app`. - The text shows only when `herdr.model_app` equals the agent's reported `app`. When an agent exits or another starts in that terminal, its old values stop showing. - Text updates live when labels change. There are no new clicks or keys. - A reporter writes the labels. For Claude Code, the status line command gets `model.display_name` and `effort.level` on stdin and runs `hc label "$HC_TERMINAL_ID" ...`. It prints nothing, and exits at once if `HC_TERMINAL_ID` is unset, so it does nothing outside the desktop app. ## Built from - Layout: a clone of `main` with `content.mod` changed in two places: the `agents` block to `model-effort-agents`, and the pane header to `model-effort-pane`. - Mods, each from `mod.clone` of the shipped mod, with `mod.json` `{ "place": "none" }`: - `model-effort-pane`, from `pane`. Adds `terminals` to its imports, and a child after the name. - `model-effort-agents`, from `agents`. Adds `.tooltip(text)` to both row kinds, using the `terminals()` list it already has. - Both add this function: ```js function modelText(labels, app) { if (!labels?.["herdr.model"] || labels["herdr.model_app"] !== app) return ""; return [labels["herdr.model"], labels["herdr.effort"]].filter(Boolean).join(" · "); } ``` - `herdr` functions: `panes()`, `agents()`, `terminals()`. - Claude Code settings: `"statusLine": { "type": "command", "command": "sh ~/.claude/statusline-herdr.sh" }`. - The mods run no programs and make no network requests. The reporter runs `jq` and `hc`. No art beyond the shipped icons. ## Watch out - `agents()` and `panes()` don't carry labels. Find the terminal in `terminals()`: `address` matches `agents()[n].address` and `panes()[n].terminal`. - `hc label` removes a label given an empty value, so writing `herdr.effort=` is correct for models with no effort. - An empty status line still takes up a row in Claude Code. - `/model` and `/effort` aren't documented status line triggers, so changes can lag until the next update. - Labels are terminal state: without the `herdr.model_app` check, they would outlive the agent. - Not done: other agents, context use, cost, or rate limits. ## Done when - `herdr-app check` passes for both mods, and `herdr-app status` shows no errors. - `hc get ` shows all three labels for a running Claude Code session. - That pane's header shows the model and effort, and hovering its agent row shows the same. - After `hc label herdr.model_app=other`, the header text disappears. It comes back after setting it to `claude-code`.