castle-web-cli 0.4.12 → 0.4.14

package/dist/agent-prompts.js

@@ -11,10 +11,8 @@ const ROUTER_RULES = `You are Castle's create assistant: the fast conversational
 Hard rules:
 - You NEVER edit files or run state-changing commands. All building and fixing happens through background task agents -- always hand the longer work to them.
 - You are the fast lane: get to your final reply as quickly as possible. When the user reports something broken, do NOT dig into the code to diagnose it first -- spawn a task whose job is to investigate AND fix it. Only read deck files when your reply itself needs them (answering a question about the deck, grounding a claim -- never make things up); never read as pre-work before spawning a task.
-- Decompose with judgment -- parallel tasks are powerful but add overhead. Parallelize when the work is ambitious enough to be worth it, not for small asks.
-- For a new game (or a substantial new part of one): make sure the FIRST task delivers a small coherent playable demo on its own, so the user is not waiting on several tasks before playing anything. Then fan out in parallel to take it further.
-- Work on ART and MECHANICS simultaneously as separate tasks, with a later task (\`after:\` both) that combines them.
-- Parallel tasks must each be separately testable and must not step on each other's files.
+- ALWAYS break the user's request into separately testable, parallelizable chunks: several small focused tasks over one big one, each something the user can try in the running deck on its own.
+- Write task prompts for speed: ask each task for the quickest viable, testable change that still delivers a meaningful step up (one or a few features, pieces of art, etc.). Only tell a task to take its time and dive deep when the user specifically asked for that.
 - To spawn a background task, include a fenced block in your reply:
 
 \`\`\`castle-task
@@ -27,6 +25,11 @@ what to build or fix, which files matter, and what "done" looks like.
 
 - Use \`after:\` only when a task truly builds on or would conflict with another (it may reference tasks spawned in this same reply, by title). Independent tasks must NOT wait on each other.
 - Use \`supersedes:\` whenever the new task fixes, redoes, or makes obsolete an earlier task -- the old row is checked off the user's board automatically. Keep the board meaning "what to look at right now".
+- When the user confirms something works (or tells you to clear items), check those finished tasks off the board by including:
+
+\`\`\`castle-done
+comma-separated titles or ids of the finished tasks
+\`\`\`
 - Tasks are one-and-done -- when the user gives feedback on a finished task, spawn a new fix task rather than reopening the old one.
 - Task agents are capable coding agents working in this same deck directory, but they know nothing about this conversation beyond your prompt.
 
@@ -92,6 +95,7 @@ Your task (id ${opts.taskId}): ${opts.title}
 ${opts.prompt}${deps}
 
 Operating rules:
+- Speed is of the essence -- go for the quickest viable, testable change, unless this prompt explicitly tells you to take longer and dive deep. It must still amount to at least one meaningful step up (one or a few features, pieces of art, etc.), never a token gesture.
 - Do this one task completely, then stop. Do not expand scope.
 - Update your progress VERY frequently: write a bare integer 0-100 to ${opts.progressPath} (e.g. \`echo 30 > ${opts.progressPath}\`) every time you advance -- at least every 10 points, or every 20 for properly small tasks. Start near 10, write 90 just before wrapping up. Never let it sit stale while you work.
 - Before finishing, write ${opts.notesPath}: a SHORT test guide for the user -- 2-4 brief sentences. Lead with exactly what to try in the running deck; mention a blocker or open question if you hit one. NO file-by-file implementation detail, no code names unless the user needs them to test. The user reads this verbatim when checking your work off.

package/dist/agent.js

@@ -70,7 +70,7 @@ const ROUTER_TIMEOUT_MS = 3 * 60_000;
 const TASK_TIMEOUT_MS = 30 * 60_000;
 const MAX_TASK_ATTEMPTS = 3;
 const TASK_POLL_MS = 1_000;
-const TASK_FENCE = '```castle-task';
+const FENCE_HOLDBACK = '```castle-';
 const RESULT_SUMMARY_CHARS = 600;
 const MAX_ATTACHMENTS = 6;
 const MAX_ATTACHMENT_BYTES = 8 * 1024 * 1024;
@@ -94,11 +94,11 @@ function readJsonFile(filePath) {
 // marker at the tail so "```cast..." never flashes up before we know what it
 // is. The full cleaned text replaces the streamed text when the reply ends.
 function visibleLength(raw) {
-    const idx = raw.indexOf(TASK_FENCE);
+    const idx = raw.indexOf(FENCE_HOLDBACK);
     if (idx >= 0)
         return idx;
-    for (let k = TASK_FENCE.length - 1; k > 0; k--) {
-        if (raw.endsWith(TASK_FENCE.slice(0, k)))
+    for (let k = FENCE_HOLDBACK.length - 1; k > 0; k--) {
+        if (raw.endsWith(FENCE_HOLDBACK.slice(0, k)))
             return raw.length - k;
     }
     return raw.length;
@@ -108,8 +108,18 @@ function visibleLength(raw) {
 // either order), then the task prompt.
 function extractDirectives(full) {
     const directives = [];
+    const checkoffs = [];
+    const doneRe = /```castle-done[ \t]*\r?\n([\s\S]*?)```/g;
+    const withoutDone = full.replace(doneRe, (_match, body) => {
+        for (const token of String(body).split(/[,\n]/)) {
+            const trimmed = token.trim();
+            if (trimmed)
+                checkoffs.push(trimmed);
+        }
+        return '';
+    });
     const fenceRe = /```castle-task[ \t]*\r?\n([\s\S]*?)```/g;
-    const cleaned = full.replace(fenceRe, (_match, body) => {
+    const cleaned = withoutDone.replace(fenceRe, (_match, body) => {
         const lines = String(body).replace(/\r/g, '').split('\n');
         const title = (lines.shift() ?? '').trim();
         const headers = { after: [], supersedes: [] };
@@ -129,7 +139,7 @@ function extractDirectives(full) {
         }
         return '';
     });
-    return { cleaned: cleaned.replace(/\n{3,}/g, '\n\n').trim(), directives };
+    return { cleaned: cleaned.replace(/\n{3,}/g, '\n\n').trim(), directives, checkoffs };
 }
 // Claude names tools directly (Read, Edit, Bash, ...).
 function claudeToolActivityLabel(name) {
@@ -412,7 +422,7 @@ function createTaskStore(opts) {
         });
     }
     function maybeStart(task) {
-        if (task.status !== 'waiting' || !depsAreSettled(task))
+        if (task.status !== 'waiting' || task.acknowledged || !depsAreSettled(task))
             return;
         start(task);
     }
@@ -434,6 +444,18 @@ function createTaskStore(opts) {
         // equivalent slash command.
         const invocation = buildAgentInvocation(backend, 'task', backend === 'claude' ? `/goal ${taskPrompt}` : taskPrompt, opts.claudeModel());
         let result = { ok: false, finalText: '', error: 'not run' };
+        let lineBuf = '';
+        const flushFeedLines = (delta) => {
+            lineBuf += delta;
+            let nl = lineBuf.indexOf('\n');
+            while (nl >= 0) {
+                const line = lineBuf.slice(0, nl).trim();
+                lineBuf = lineBuf.slice(nl + 1);
+                if (line)
+                    opts.onFeed(task, line);
+                nl = lineBuf.indexOf('\n');
+            }
+        };
         for (let attempt = 1; attempt <= MAX_TASK_ATTEMPTS; attempt++) {
             result = await runAgentCli({
                 cwd: deckDir,
@@ -446,6 +468,11 @@ function createTaskStore(opts) {
                 onSpawn: (pid) => {
                     task.pid = pid;
                 },
+                onDelta: (delta) => flushFeedLines(delta),
+                onActivity: (activity) => {
+                    if (activity)
+                        opts.onFeed(task, `[${activity}]`);
+                },
             });
             if (!result.crashed)
                 return result;
@@ -480,10 +507,13 @@ function createTaskStore(opts) {
         });
     }
     function spawnFromDirective(directive, originMessageId) {
-        // A fix/redo task sweeps the rows it obsoletes off the user's board.
+        // A fix/redo task sweeps the rows it obsoletes off the user's board. A
+        // superseded task that never started must not start later either.
         for (const id of resolveDeps(tasks, directive.supersedes)) {
             const old = tasks.get(id);
             if (old && !old.acknowledged) {
+                if (old.status === 'waiting')
+                    old.status = 'interrupted';
                 old.acknowledged = true;
                 touch(old);
                 opts.onSuperseded(old);
@@ -536,7 +566,12 @@ function createTaskStore(opts) {
             }
         }
     }
-    return { sorted, get: (id) => tasks.get(id), spawnFromDirective, acknowledge, shutdown };
+    // The router checks finished tasks off by title or id (castle-done fence).
+    function checkOff(tokens) {
+        for (const id of resolveDeps(tasks, tokens))
+            acknowledge(id, false);
+    }
+    return { sorted, get: (id) => tasks.get(id), spawnFromDirective, acknowledge, checkOff, shutdown };
 }
 // -- attachments ----------------------------------------------------------------
 const ATTACHMENT_MIME = {
@@ -581,6 +616,23 @@ function asPromptTask(task) {
         notes: task.notes,
     };
 }
+function createTaskFeeds(broadcast) {
+    const map = new Map();
+    function push(task, entry) {
+        let feed = map.get(task.id);
+        if (!feed) {
+            feed = [];
+            map.set(task.id, feed);
+        }
+        if (feed[feed.length - 1] === entry)
+            return;
+        feed.push(entry);
+        if (feed.length > 80)
+            feed.splice(0, feed.length - 80);
+        broadcast({ type: 'task-feed', id: task.id, entry });
+    }
+    return { map, push };
+}
 function createMessageLog(messagesPath, broadcast) {
     const loaded = readJsonFile(messagesPath) ?? [];
     const messages = loaded
@@ -695,7 +747,9 @@ function runRouterTurnIn(ctx, instruction) {
             });
             return;
         }
-        const { cleaned, directives } = extractDirectives(result.finalText);
+        const { cleaned, directives, checkoffs } = extractDirectives(result.finalText);
+        if (result.ok && checkoffs.length > 0)
+            ctx.taskStore.checkOff(checkoffs);
         // Drop directives from stale turns, and any whose title matches a task
         // already in flight (two runs reacting to the same ask).
         const stale = epoch !== ctx.currentEpoch();
@@ -721,6 +775,27 @@ function runRouterTurnIn(ctx, instruction) {
         });
     });
 }
+// Merge settings changes from the client: validate, persist, broadcast, log.
+function applyAgentSettings(incoming, ctx) {
+    const { settings } = ctx;
+    const changes = [];
+    for (const key of ['router', 'tasks']) {
+        const value = normalizeBackend(incoming[key]);
+        if (value && value !== settings[key]) {
+            settings[key] = value;
+            changes.push(`${key} agent -> ${value}`);
+        }
+    }
+    const model = normalizeClaudeModel(incoming.claudeModel);
+    if (model && model !== settings.claudeModel) {
+        settings.claudeModel = model;
+        changes.push(`claude model -> ${model}`);
+    }
+    if (changes.length === 0)
+        return;
+    fs.writeFileSync(ctx.settingsPath, JSON.stringify(settings, null, 2) + '\n');
+    ctx.broadcast({ type: 'settings', settings });
+}
 export function createAgentServer(opts) {
     const { deckDir, deckLabel } = opts;
     const agentDir = path.join(deckDir, '.castle', 'agent');
@@ -750,27 +825,8 @@ export function createAgentServer(opts) {
         tasks: normalizeBackend(storedSettings?.tasks) ?? DEFAULT_SETTINGS.tasks,
         claudeModel: normalizeClaudeModel(storedSettings?.claudeModel) ?? DEFAULT_SETTINGS.claudeModel,
     };
-    function applySettings(incoming) {
-        const changes = [];
-        for (const key of ['router', 'tasks']) {
-            const value = normalizeBackend(incoming[key]);
-            if (value && value !== settings[key]) {
-                settings[key] = value;
-                changes.push(`${key} agent -> ${value}`);
-            }
-        }
-        const model = normalizeClaudeModel(incoming.claudeModel);
-        if (model && model !== settings.claudeModel) {
-            settings.claudeModel = model;
-            changes.push(`claude model -> ${model}`);
-        }
-        if (changes.length === 0)
-            return;
-        fs.writeFileSync(settingsPath, JSON.stringify(settings, null, 2) + '\n');
-        broadcast({ type: 'settings', settings });
-        for (const change of changes)
-            addLog(`settings: ${change}`);
-    }
+    const applySettings = (incoming) => applyAgentSettings(incoming, { settings, settingsPath, broadcast });
+    const taskFeeds = createTaskFeeds(broadcast);
     const taskStore = createTaskStore({
         deckDir,
         deckLabel,
@@ -782,8 +838,9 @@ export function createAgentServer(opts) {
         onUpdate: (task) => broadcast({ type: 'task-update', task }),
         onStarted: () => undefined,
         onRetry: (task, attempt) => addLog(`agent died, retrying (${attempt}/${MAX_TASK_ATTEMPTS}): ${task.title}`),
-        onFinished: () => undefined,
+        onFinished: (task) => taskFeeds.map.delete(task.id),
         onSuperseded: () => undefined,
+        onFeed: (task, entry) => taskFeeds.push(task, entry),
     });
     // A new user message interrupts the in-flight router reply: its partial
     // text stays in the log, and the next turn continues both threads. The
@@ -845,7 +902,8 @@ export function createAgentServer(opts) {
     const wss = new WebSocketServer({ noServer: true });
     function attachClient(socket) {
         clients.add(socket);
-        socket.send(JSON.stringify({ type: 'hello', messages, tasks: taskStore.sorted(), settings }));
+        const hello = { type: 'hello', messages, tasks: taskStore.sorted(), settings, feeds: Object.fromEntries(taskFeeds.map) };
+        socket.send(JSON.stringify(hello));
         socket.on('message', (rawData) => {
             let msg;
             try {

package/dist/chat-client.js

@@ -136,15 +136,22 @@ function initPanelChrome() {
 //
 // React app.
 //
-function applyServerEvent(ev, setMessages, setTasks) {
+function applyServerEvent(ev, setMessages, setTasks, setFeeds) {
     if (ev.type === 'hello') {
         const messages = ev.messages ?? [];
         const tasks = ev.tasks ?? [];
+        const feeds = ev.feeds ?? {};
         setMessages(() => messages);
         setTasks(() => tasks);
+        setFeeds(() => feeds);
         if (ev.settings)
             applyBackendSegs(ev.settings);
     }
+    else if (ev.type === 'task-feed' && ev.id) {
+        const id = ev.id;
+        const entry = ev.entry ?? '';
+        setFeeds((prev) => ({ ...prev, [id]: [...(prev[id] ?? []).slice(-79), entry] }));
+    }
     else if (ev.type === 'settings' && ev.settings) {
         applyBackendSegs(ev.settings);
     }
@@ -176,17 +183,37 @@ function applyServerEvent(ev, setMessages, setTasks) {
             const next = known ? prev.map((t) => (t.id === task.id ? task : t)) : [...prev, task];
             return next.sort((a, b) => a.createdAt.localeCompare(b.createdAt));
         });
+        if (TERMINAL_TASK_STATUSES.includes(task.status)) {
+            setFeeds((prev) => {
+                if (!(task.id in prev))
+                    return prev;
+                const next = { ...prev };
+                delete next[task.id];
+                return next;
+            });
+        }
     }
 }
+function TaskFeed(props) {
+    const hostRef = React.useRef(null);
+    React.useLayoutEffect(() => {
+        const host = hostRef.current;
+        if (host)
+            host.scrollTop = host.scrollHeight;
+    }, [props.lines]);
+    return (React.createElement("div", { className: "task-feed", ref: hostRef, onClick: (event) => event.stopPropagation() }, props.lines.map((line, index) => (React.createElement("div", { key: index }, line)))));
+}
 function TaskRow(props) {
     const { task, onAck } = props;
     const [expanded, setExpanded] = React.useState(false);
-    const pct = task.status === 'done' ? 100 : task.progress;
     const finished = TERMINAL_TASK_STATUSES.includes(task.status);
+    // A full circle means "click me to check off". A still-running task never
+    // fills completely, even if its agent already reports 100.
+    const pct = task.status === 'done' ? 100 : finished ? task.progress : Math.min(task.progress, 95);
     const notes = task.notes.trim() || task.resultSummary?.trim() || '(no notes from the agent yet)';
     return (React.createElement("div", { className: `task-card${task.status === 'waiting' ? ' waiting' : ''}`, onClick: () => setExpanded(!expanded) },
         React.createElement("div", { className: "task-row" },
-            React.createElement("span", { className: `task-pie${finished ? ' task-ack-pie' : ''}`, title: finished ? 'tested -- check off' : undefined, style: { background: `conic-gradient(#0969da ${pct}%, #eaeef2 0)` }, onClick: finished
+            React.createElement("span", { className: `task-pie${finished ? ' task-ack-pie' : ''}`, title: finished ? 'tested -- check off' : undefined, style: { background: `conic-gradient(#000000 ${pct}%, #eaeef2 0)` }, onClick: finished
                     ? (event) => {
                         event.stopPropagation();
                         onAck(task.id, false);
@@ -196,13 +223,14 @@ function TaskRow(props) {
             task.status !== 'done' ? (React.createElement("span", { className: `task-status ${task.status}` }, task.status === 'running' ? `${pct}%` : task.status)) : null),
         React.createElement("div", { className: "task-bar" },
             React.createElement("div", { style: { width: `${pct}%` } })),
-        expanded ? (React.createElement("div", { className: "task-notes", dangerouslySetInnerHTML: renderMarkdown(notes) })) : null));
+        expanded && task.status === 'running' && props.feed && props.feed.length > 0 ? (React.createElement(TaskFeed, { lines: props.feed })) : null,
+        expanded && task.status !== 'running' ? (React.createElement("div", { className: "task-notes", dangerouslySetInnerHTML: renderMarkdown(notes) })) : null));
 }
 function TaskBoard(props) {
     const visible = props.tasks.filter((t) => !t.acknowledged);
     if (visible.length === 0)
         return null;
-    return (React.createElement("div", { id: "chat-strip" }, visible.map((task) => (React.createElement(TaskRow, { key: task.id, task: task, onAck: props.onAck })))));
+    return (React.createElement("div", { id: "chat-strip" }, visible.map((task) => (React.createElement(TaskRow, { key: task.id, task: task, feed: props.feeds[task.id], onAck: props.onAck })))));
 }
 function Message(props) {
     const { msg } = props;
@@ -315,6 +343,7 @@ function InputRow(props) {
 function App() {
     const [messages, setMessages] = React.useState([]);
     const [tasks, setTasks] = React.useState([]);
+    const [feeds, setFeeds] = React.useState({});
     const sendRef = React.useRef(() => undefined);
     React.useEffect(() => {
         const wsProtocol = window.location.protocol === 'https:' ? 'wss:' : 'ws:';
@@ -345,7 +374,7 @@ function App() {
             });
             sock.addEventListener('message', (event) => {
                 try {
-                    applyServerEvent(JSON.parse(String(event.data)), setMessages, setTasks);
+                    applyServerEvent(JSON.parse(String(event.data)), setMessages, setTasks, setFeeds);
                 }
                 catch {
                     /* malformed frame */
@@ -382,7 +411,7 @@ function App() {
         };
     }, []);
     return (React.createElement(React.Fragment, null,
-        React.createElement(TaskBoard, { tasks: tasks, onAck: (id, rejected) => sendRef.current({ type: 'task-ack', id, rejected }) }),
+        React.createElement(TaskBoard, { tasks: tasks, feeds: feeds, onAck: (id, rejected) => sendRef.current({ type: 'task-ack', id, rejected }) }),
         React.createElement(MessageList, { messages: messages }),
         React.createElement(InputRow, { onSend: (text, images) => sendRef.current({ type: 'user-message', text, images }) })));
 }

package/dist/ide.js

@@ -626,6 +626,32 @@ const IDE_STYLES = `
     word-break: break-word;
     font-size: 14px;
   }
+  /* Live stream of a running task agent (text lines + [tool labels]) shown
+     on expand -- capped height, auto-scrolled, fading at both scroll edges. */
+  .task-feed {
+    margin-top: 6px;
+    max-height: 150px;
+    overflow-y: auto;
+    font-size: 12px;
+    line-height: 1.5;
+    color: #57606a;
+    cursor: default;
+    word-break: break-word;
+    -webkit-mask-image: linear-gradient(
+      to bottom,
+      transparent,
+      #000000 14px,
+      #000000 calc(100% - 14px),
+      transparent
+    );
+    mask-image: linear-gradient(
+      to bottom,
+      transparent,
+      #000000 14px,
+      #000000 calc(100% - 14px),
+      transparent
+    );
+  }
   .task-row { display: flex; align-items: center; gap: 8px; }
   .task-title {
     flex: 1 1 auto;
@@ -634,6 +660,7 @@ const IDE_STYLES = `
     text-overflow: ellipsis;
     white-space: nowrap;
   }
+  .task-title::first-letter { text-transform: uppercase; }
   .task-status { color: #6e7781; font-size: 12px; flex: 0 0 auto; }
   .task-status.failed, .task-status.interrupted { color: #a40e26; }
   .task-pie {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "castle-web-cli",
-  "version": "0.4.12",
+  "version": "0.4.14",
   "type": "module",
   "bin": {
     "castle-web": "./dist/index.js"