[{"data":1,"prerenderedAt":28},["ShallowReactive",2],{"doc:en:overview":3},{"title":4,"lede":5,"html":6,"headings":7},"Keeping track of your agents","When several agents work at the same time, you need to know which one is waiting for what. This page presents the overview, which lists all agents of all projects, the shortcuts for moving from one agent to another, and the search in conversations.","\u003Cp>On macOS, \u003Ccode class=\"key\">Ctrl\u003C\u002Fcode> becomes \u003Ccode class=\"key\">⌘\u003C\u002Fcode> for shortcuts (except \u003Ccode class=\"key\">Ctrl+Tab\u003C\u002Fcode>).\u003C\u002Fp>\n\u003Ch2 id=\"the-overview\">The overview\u003C\u002Fh2>\n\u003Cp>The overview takes the place of the project on screen to list all agents of all projects. Open it with the “Overview” button in the title bar, to the left of the project tabs, or with \u003Ccode class=\"key\">Ctrl+Shift+A\u003C\u002Fcode> (the shortcut also works from a terminal). The same button or shortcut closes it; so does \u003Ccode class=\"key\">Esc\u003C\u002Fcode>, and you’re back to the view it was hiding: the project or the statistics.\u003C\u002Fp>\n\u003Ch3 id=\"what-it-shows\">What it shows\u003C\u002Fh3>\n\u003Cp>The header gives the count: “2 projects · 7 agents · 3 running”. Below, the agents are arranged in groups:\u003C\u002Fp>\n\u003Col>\n\u003Cli>\u003Cstrong>“Waiting for your answer”\u003C\u002Fstrong>: the agents with the “Question” status, from all projects, the one that has waited longest first, with its project’s name.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>One group per project\u003C\u002Fstrong>, in the order of the tabs, with its color; the agents are in the order of the sidebar.\u003C\u002Fli>\n\u003C\u002Fol>\n\u003Cp>An agent appears only once, and archived agents aren’t listed. Each line has six columns:\u003C\u002Fp>\n\u003Cdiv class=\"table\">\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Column\u003C\u002Fth>\n\u003Cth>Content\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\u003Ctr>\n\u003Ctd>Agent\u003C\u002Ftd>\n\u003Ctd>The status dot, the name, and the ticket’s \u003Ccode>▸ KEY\u003C\u002Fcode> label if there is one\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Activity\u003C\u002Ftd>\n\u003Ctd>“Question”, “Resumes at 3:00 PM” (usage limit), “Setting up the worktree” followed by the step, or what the agent is doing (“Reading src\u002Fa.ts”, “Thinking”…); for an agent at rest, its status: “Ready”, “Done”, or “Error”\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Model\u003C\u002Ftd>\n\u003Ctd>The agent’s model\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Cost\u003C\u002Ftd>\n\u003Ctd>The agent’s cost, preceded by “≈” during a turn\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Context\u003C\u002Ftd>\n\u003Ctd>How much of the model’s window is already filled, or “—” while it isn’t known\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>Since\u003C\u002Ftd>\n\u003Ctd>The time elapsed since its last change (“Last change” with the time in the tooltip)\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Cp>If the agent publishes a progress line through Escouade’s MCP server (see \u003Ca href=\"\u002Fen\u002Fdocs\u002Fmcp\u002F\">the MCP server\u003C\u002Fa>), it’s written under its line.\u003C\u002Fp>\n\u003Cp>A click on a line, or \u003Ccode class=\"key\">Enter\u003C\u002Fcode> when it has focus, opens the agent in its project, on its conversation.\u003C\u002Fp>\n\u003Ch3 id=\"answering-on-the-spot\">Answering on the spot\u003C\u002Fh3>\n\u003Cp>The lines of the “Waiting for your answer” group show the agent’s request, and you can answer some of them without leaving the view.\u003C\u002Fp>\n\u003Cul>\n\u003Cli>A \u003Cstrong>permission request\u003C\u002Fstrong> that can be read in full here (a command, a file, a search, an address: the \u003Ccode class=\"key\">Bash\u003C\u002Fcode>, \u003Ccode class=\"key\">PowerShell\u003C\u002Fcode>, \u003Ccode class=\"key\">SlashCommand\u003C\u002Fcode>, \u003Ccode class=\"key\">Read\u003C\u002Fcode>, \u003Ccode class=\"key\">Edit\u003C\u002Fcode>, \u003Ccode class=\"key\">Write\u003C\u002Fcode>, \u003Ccode class=\"key\">MultiEdit\u003C\u002Fcode>, \u003Ccode class=\"key\">NotebookEdit\u003C\u002Fcode>, \u003Ccode class=\"key\">Grep\u003C\u002Fcode>, \u003Ccode class=\"key\">Glob\u003C\u002Fcode>, \u003Ccode class=\"key\">WebFetch\u003C\u002Fcode>, and \u003Ccode class=\"key\">WebSearch\u003C\u002Fcode> tools) shows the tool and what it wants to do, with “Allow” and “Deny” buttons. There’s no “Always allow”: for that, or to deny with an explanation, open the conversation.\u003C\u002Fli>\n\u003Cli>For everything else, the line shows the request and a “Reply” button that opens the agent: a \u003Cstrong>question\u003C\u002Fstrong> (its text is shown), a \u003Cstrong>plan\u003C\u002Fstrong> (“Claude suggests a plan”), an agent waiting in the Claude Code view (“Waiting for your answer in Claude Code’s interface”), a request that Claude Code would deny by default, or a tool whose summary doesn’t say everything it asks (“Read it in the conversation before answering.”). A request whose argument runs past four lines is also sent back to the conversation (“Too long to read here: read it and answer in the conversation.”).\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Only the first request of each agent is shown; when you answer it, the next one takes its place. When a line changes, or slides because a line above it disappeared, its buttons stay inactive for half a second: a click meant for the previous request doesn’t answer the next one.\u003C\u002Fp>\n\u003Ch3 id=\"with-the-keyboard\">With the keyboard\u003C\u002Fh3>\n\u003Cdiv class=\"table\">\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Keys\u003C\u002Fth>\n\u003Cth>Effect\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\u003Ctr>\n\u003Ctd>\u003Ccode class=\"key\">↑\u003C\u002Fcode> \u002F \u003Ccode class=\"key\">↓\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>Previous \u002F next line, from one group to the next\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode class=\"key\">Home\u003C\u002Fcode> \u002F \u003Ccode class=\"key\">End\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>First \u002F last line\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode class=\"key\">Enter\u003C\u002Fcode> or \u003Ccode class=\"key\">Space\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>Opens the line’s agent\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode class=\"key\">Esc\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>Closes the overview\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Ch2 id=\"moving-between-agents\">Moving between agents\u003C\u002Fh2>\n\u003Cdiv class=\"table\">\u003Ctable>\n\u003Cthead>\n\u003Ctr>\n\u003Cth>Keys\u003C\u002Fth>\n\u003Cth>Effect\u003C\u002Fth>\n\u003C\u002Ftr>\n\u003C\u002Fthead>\n\u003Ctbody>\u003Ctr>\n\u003Ctd>\u003Ccode class=\"key\">Ctrl+J\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>Goes to the next agent that is waiting for an answer or that needs a look (a question, an end, or an error you haven’t seen yet), in all projects, starting with the one that has waited longest. The cursor is in its message field.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode class=\"key\">Ctrl+Tab\u003C\u002Fcode> \u002F \u003Ccode class=\"key\">Ctrl+Shift+Tab\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>Next \u002F previous agent of the project on screen, in sidebar order, looping around and skipping archived agents. On macOS, it really is \u003Ccode class=\"key\">Ctrl+Tab\u003C\u002Fcode>, not \u003Ccode class=\"key\">⌘Tab\u003C\u002Fcode>.\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003Ctr>\n\u003Ctd>\u003Ccode class=\"key\">Ctrl+1\u003C\u002Fcode> … \u003Ccode class=\"key\">Ctrl+9\u003C\u002Fcode>\u003C\u002Ftd>\n\u003Ctd>Goes to project n (see \u003Ca href=\"\u002Fen\u002Fdocs\u002Finterface\u002F\">the interface\u003C\u002Fa>).\u003C\u002Ftd>\n\u003C\u002Ftr>\n\u003C\u002Ftbody>\u003C\u002Ftable>\u003C\u002Fdiv>\n\u003Cp>\u003Ccode class=\"key\">Ctrl+J\u003C\u002Fcode> and \u003Ccode class=\"key\">Ctrl+Tab\u003C\u002Fcode> also work from a terminal. The “N waiting” button in the status bar, at the bottom, does the same as \u003Ccode class=\"key\">Ctrl+J\u003C\u002Fcode>.\u003C\u002Fp>\n\u003Cp>To keep an eye on the other projects without opening them, look at their tabs: a pill gives the number of agents waiting for an answer, a green dot shows that an agent is working, “Δ 7” counts the changed files, and a tab that blinks amber, green, or red flags an agent that needs a look (see \u003Ca href=\"\u002Fen\u002Fdocs\u002Fnotifications\u002F\">notifications\u003C\u002Fa>). The status bar counts the “active”, “waiting”, and “done” agents.\u003C\u002Fp>\n\u003Ch2 id=\"searching-conversations\">Searching conversations\u003C\u002Fh2>\n\u003Cp>\u003Ccode class=\"key\">Ctrl+K\u003C\u002Fcode> opens “Search conversations”, a palette that searches everything you and your agents have written. (In a terminal on Windows and Linux, \u003Ccode class=\"key\">Ctrl+K\u003C\u002Fcode> stays with the shell.)\u003C\u002Fp>\n\u003Cul>\n\u003Cli>\u003Cstrong>What is searched\u003C\u002Fstrong>: your messages and Claude’s, the commands and files of tool calls with their output, questions and their answers, permission requests, plans, notices, turn errors, and the names of attached files. Thinking, folded in the conversation, isn’t searched.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>How\u003C\u002Fstrong>: the search ignores case and accents (“cafe” finds “café”), treats \u003Ccode class=\"key\">\\\u003C\u002Fcode> and \u003Ccode class=\"key\">\u002F\u003C\u002Fcode> as equivalent in a path, and reads a typographic apostrophe as a straight one. It starts after two characters, a moment after your last keystroke, and reads the conversation logs on disk, including those of agents in the Claude Code view.\u003C\u002Fli>\n\u003Cli>\u003Cstrong>Where\u003C\u002Fstrong>: “This project” or “All projects” (the first while a project is on screen), and the “Archived agents” checkbox, checked by default.\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Results are grouped by agent, the most recently active first, with the project’s name and the “archived” label where relevant. Each result shows an excerpt with the match highlighted, and its age (“3 h ago”). The most recent messages of each agent come first.\u003C\u002Fp>\n\u003Cp>\u003Ccode class=\"key\">↑\u003C\u002Fcode> and \u003Ccode class=\"key\">↓\u003C\u002Fcode> choose a result, \u003Ccode class=\"key\">Enter\u003C\u002Fcode> or a click opens it, \u003Ccode class=\"key\">Esc\u003C\u002Fcode> closes the palette. The agent opens on its conversation (the editor, if it was open, makes room), scrolled to the message, which is highlighted for two seconds; if the message is old, the items before it are drawn first. An archived agent opens read-only. If the agent was deleted in the meantime, a message says “This agent no longer exists.”\u003C\u002Fp>\n\u003Cp>The palette stops at 200 results (“The first 200 results: narrow your search.”) or after 5 seconds of reading (“… search stopped after 5 s: narrow your search.”): add words to tighten it.\u003C\u002Fp>\n",[8,12,16,19,22,25],{"id":9,"text":10,"depth":11},"the-overview","The overview",2,{"id":13,"text":14,"depth":15},"what-it-shows","What it shows",3,{"id":17,"text":18,"depth":15},"answering-on-the-spot","Answering on the spot",{"id":20,"text":21,"depth":15},"with-the-keyboard","With the keyboard",{"id":23,"text":24,"depth":11},"moving-between-agents","Moving between agents",{"id":26,"text":27,"depth":11},"searching-conversations","Searching conversations",1791926756188]