symbiot 0.43.0 → 0.43.1

package/README.md

@@ -67,6 +67,8 @@ Tabs:
   any tab, and once you watch something the app opens on the Dashboard instead of
   the Map. `symbiot watch board` prints the same cards as JSON, and
   `symbiot watch board --line` as one line for a status bar (`2 emails · 1 WhatsApp message`).
+  `symbiot watch chat <id> "question"` is the card's 💬 chat from a terminal, so an
+  agent can go over what's new with you too.
 - **Drift** — the [`symbiot drift`](#whats-out-of-sync-symbiot-drift) report, with
   a "fetch latest" toggle (and an [experimental](#experimental) "check CI").
 - **Week / Standup / Todo** — the write-ups (these use your chosen AI).
@@ -381,7 +383,15 @@ Reading both costs about 40 lines.
 ```
 
 Up to five questions show on that agent's block in the **Agents** tab, each with
-its options and room for your own answer, alongside its ideas. Each option says
+two options and room for your own answer, alongside its ideas, two at a time.
+Two, because most people pick the recommended option or add every idea without
+weighing them, so the agent is told to be the judge. Both options have to be good
+routes, each saying in plain words what it changes for the project, the people on
+it and the company. The recommended one has to be the best for, in that order,
+the company, the people doing the work and the goal, based on evidence the agent
+checked (git history, tests, logs). An idea that's in your Tasks leaves the list,
+and the next one moves up. **Suggest next steps** on the Map follows the same rule
+and suggests two. Each option says
 who acts: **You** (a setting, a click, a command) or **Agent** (picking it is
 enough). A question that needs a release ("Once 0.41.0 is installed: …") shows
 the version installed here and the one on npm, and holds back a "Done" answer
@@ -401,8 +411,10 @@ committed. The questions stay on the Agents tab after Symbiot restarts (or when
 the answer and reminds you the step is still yours to do. Your agent waits for it
 instead of starting straight away and asking the same thing again. If the step
 names a file in backticks (`` `.env` ``), the agent starts by itself once that
-file changes, while the app runs. Otherwise **Start it now** on that folder's
-block starts it once you've done the step.
+file changes, while the app runs. A step in Settings (**Allow command**, a
+connector) changes the agent command rather than a file, so a changed agent
+command counts as the step being done too. Otherwise **Start it now** on that
+folder's block starts it once you've done the step.
 
 **A run that stopped on questions isn't repeated for nothing.** When a run ends
 having asked questions you haven't answered, with tasks still unticked, Symbiot
@@ -770,6 +782,7 @@ symbiot watch new --hours 24               # what's new, newest first (JSON)
 symbiot watch board                        # the Dashboard's cards (JSON; .total is the count, for a status bar)
 symbiot watch board --line                 # the same as one line: "2 emails · 1 WhatsApp message" (empty when nothing's new)
 symbiot watch seen <watch id>              # set a card back to 0, like its Seen button
+symbiot watch chat <watch id> "what needs me?"  # talk a card over with your AI, like its 💬 (--clear starts over)
 symbiot watch check                        # read them all now (JSON)
 symbiot watch draft <id>                   # Draft a reply to a new email or chat message (see below)
 symbiot watch brief on                     # your AI says what needs you (off to stop)

package/agents.mjs

@@ -112,9 +112,9 @@ function runHandoff(repoPath, { force = false } = {}) {
   const busy = runningHandoff(repoPath); if (busy) return { busy: true, id: busy.id || "", pid: busy.pid, auto: !!busy.auto };
   releaseHeldTasks(repoPath); // held for an agent another process started, which has since exited
   tmpl = withConnectors(tmpl, repoPath);
-  if (force) { try { unlinkSync(join(repoPath, ".symbiot", WAITING)); } catch {} }
-  else { const w = waitingFor(repoPath); if (w) return { blocked: true, waiting: true, questions: 0, note: waitNote(w) }; }
+  if (!force) { const w = waitingFor(repoPath); if (w) return { blocked: true, waiting: true, questions: 0, note: waitNote(w) }; }
   const blocked = !force && blockedAgain(repoPath, tmpl); if (blocked) return blocked;
+  try { unlinkSync(join(repoPath, ".symbiot", WAITING)); } catch {} // the step's done (or this is Start it anyway): this is the run it waited for
   const lock = join(repoPath, ".symbiot", LOCK);
   const env = opts.env && typeof opts.env === "object" ? Object.fromEntries(Object.entries(opts.env).map(([k, v]) => [k, String(v)])) : null;
   const e = track(typeof opts.name === "string" && opts.name ? opts.name.slice(0, 80) : repoPath.split("/").pop(), fillHandoff(tmpl, repoPath), repoPath, (code) => {
@@ -165,11 +165,17 @@ function earlierRuns() {
 // only gets the same questions back. So answerQuestions reminds them the step is
 // theirs, and Send answers & continue waits instead of starting: .symbiot/
 // waiting.json keeps the step and how each file it names (in backticks, like
-// `.env`) was then. Once one of them changes, the next run goes ahead
-// (startWaiting starts the one that was asked for); with no file to watch, it
+// `.env`) was then, and the agent command: a step done in Settings (Allow
+// command, a connector) changes that, not a file, so it counts as done too, as
+// it does for blocked.json. Once one of them changes, the next run goes ahead
+// (startWaiting starts the one that was asked for); with neither changed, it
 // waits for Start it now. Start it anyway (force) always goes ahead.
 const WAITING = "waiting.json";
 const yourStep = (a) => /👤/.test(String(a || ""));
+// A step in Settings → Handoff: the note names the command as what it waits on.
+const settingsStep = (s) => /\bsettings\b|\ballow\b|\bgrant|connector/i.test(String(s || ""));
+// The command the folder's next run would get, connectors and all (runHandoff).
+const runCmd = (path) => digest(withConnectors(handoffCmd(), path));
 // The files a step names, in backticks: `.env`, `~/.termux/termux.properties`,
 // `.claude/settings.json`. Not a command (`npm test`), a site (`web.whatsapp.com`,
 // unless there's a file by that name) or a path that can't be one here.
@@ -186,27 +192,36 @@ function namedFiles(text, dir) {
   return out.filter((f, i) => out.findIndex((g) => g.path === f.path) === i).slice(0, 5);
 }
 const fileSig = (p) => { try { const s = statSync(p); return s.mtimeMs + ":" + s.size; } catch { return "none"; } };
-// What a folder is waiting on: { step, files: [names], at, rerun }, or null (and
-// waiting.json goes) once a file it names has changed.
+// What a folder is waiting on: { step, files: [names], cmd, at, rerun }, or null
+// once a file it names, or the agent command, has changed. Then waiting.json
+// goes, unless a run was asked for: startWaiting needs it to start that one (the
+// Agents tab asks this every few seconds, and used to lose it first), and the
+// run deletes it as it starts. cmd: the step is one in Settings, so the note says
+// the command counts.
 function waitingFor(path) {
   let w = null; try { w = JSON.parse(readSymbiot(path, WAITING)); } catch {}
   if (!w || typeof w.step !== "string") return null;
   const files = Array.isArray(w.files) ? w.files : [];
-  if (files.some((f) => f && fileSig(f.path) !== f.sig)) { try { unlinkSync(join(path, ".symbiot", WAITING)); } catch {} return null; }
-  return { step: w.step, files: files.map((f) => f.name), at: w.at || 0, rerun: !!w.rerun };
+  if (files.some((f) => f && fileSig(f.path) !== f.sig) || (w.cmd && runCmd(path) !== w.cmd)) { if (!w.rerun) try { unlinkSync(join(path, ".symbiot", WAITING)); } catch {} return null; }
+  return { step: w.step, files: files.map((f) => f.name), cmd: !!w.cmd && settingsStep(w.step), at: w.at || 0, rerun: !!w.rerun };
+}
+// "`.env` or the agent command in Settings → Handoff", what a wait ends on ("" if neither).
+function waitOn(files, cmd) {
+  const names = files.map((f) => "`" + f + "`").join(" or ");
+  return cmd ? (names ? names + " or " : "") + "the agent command (Settings → Handoff)" : names;
 }
 function waitNote(w) {
-  const names = w.files.map((f) => "`" + f + "`").join(" or ");
-  return `Not started: your step comes first: ${w.step.replace(/^\s*👤\s*(You:)?\s*/, "")} ${names ? `Your agent starts once ${names} changes.` : "Once it's done, click Start it now (or Start it anyway)."}`;
+  const on = waitOn(w.files, w.cmd);
+  return `Not started: your step comes first: ${w.step.replace(/^\s*👤\s*(You:)?\s*/, "")} ${on ? `Your agent starts once ${on} changes. Done it some other way? Click Start it now.` : "Once it's done, click Start it now (or Start it anyway)."}`;
 }
-// Runs that waited for a file of yours, and were asked for, start once it changes.
-// The app calls this every so often. Returns the jobs started.
+// Runs that waited for a file of yours (or the agent command), and were asked
+// for, start once it changes. The app calls this every so often. Returns the jobs started.
 function startWaiting() {
   const started = [];
   for (const r of loadRuns()) {
     let w = null; try { w = JSON.parse(readSymbiot(r.path, WAITING)); } catch {}
-    if (!w || !w.rerun || !Array.isArray(w.files) || !w.files.length || runningHandoff(r.path)) continue;
-    if (waitingFor(r.path)) continue; // nothing it names has changed yet
+    if (!w || !w.rerun || !Array.isArray(w.files) || !(w.files.length || w.cmd) || runningHandoff(r.path)) continue;
+    if (waitingFor(r.path)) continue; // nothing it names, nor the command, has changed yet
     const e = runHandoff(r.path); if (e && !e.busy && !e.blocked) started.push(e);
   }
   return started;
@@ -438,6 +453,12 @@ function agentChanges(path, startedAt) {
 // .symbiot/ANSWERS.md for the next run to read. Plain files, so it works the
 // same whichever model or tool the agent is.
 const QUESTIONS_MAX = 5;
+// Two at a time: most people pick the recommended option, or + task an idea,
+// without weighing the rest, and the choices are the agent's to judge (the
+// brief says how). So a question shows its first two options, and the Agents
+// tab the ideas two at a time (`ideasShown`): one in Tasks leaves the list, and
+// the next moves up. The rest are a click away.
+const OPTIONS_SHOWN = 2, IDEAS_SHOWN = 2;
 const qKey = (s) => String(s || "").trim().toLowerCase().replace(/\s+/g, " ");
 const readSymbiot = (path, f) => { try { return readFileSync(join(path, ".symbiot", f), "utf8"); } catch { return ""; } };
 // "## Questions" → "### question", context lines, "- option" bullets;
@@ -479,9 +500,10 @@ function agentQuestions(path, repo) {
   const done = new Set([...readSymbiot(path, "ANSWERS.md").matchAll(/^###\s+(.+)$/gm)].map((m) => qKey(m[1])));
   const open = p.questions.filter((x) => !done.has(qKey(x.q)));
   const tasks = p.suggestions.length ? loadTasks() : [];
+  const ideas = p.suggestions.map(suggestionTarget).map(({ repo: to, text }) => ({ text, repo: to || repo, other: !!to && to !== repo, added: tasks.some((t) => t.repo === (to || repo) && sameTask(t.text, text)) })).filter((s) => !s.added);
   return {
-    questions: open.slice(0, QUESTIONS_MAX), answered: p.questions.length - open.length,
-    suggestions: p.suggestions.map(suggestionTarget).map(({ repo: to, text }) => ({ text, repo: to || repo, other: !!to && to !== repo, added: tasks.some((t) => t.repo === (to || repo) && sameTask(t.text, text)) })),
+    questions: open.slice(0, QUESTIONS_MAX).map((x) => ({ ...x, options: x.options.slice(0, OPTIONS_SHOWN) })), answered: p.questions.length - open.length,
+    suggestions: ideas, ideasShown: IDEAS_SHOWN,
   };
 }
 // Save answers to .symbiot/ANSWERS.md; opts.rerun hands the repo back to the
@@ -508,10 +530,10 @@ function answerQuestions(path, answers, opts = {}) {
   const yours = rows.filter((x) => yourStep(x.a)).map((x) => stepText(x.a)).filter(Boolean);
   if (yours.length) {
     const step = yours.join(" "), files = namedFiles(rows.filter((x) => yourStep(x.a)).map((x) => x.a.split("🤖")[0]).join(" "), path);
-    try { writeFileSync(join(path, ".symbiot", WAITING), JSON.stringify({ step, files: files.map((f) => ({ ...f, sig: fileSig(f.path) })), at: Date.now(), rerun: !!opts.rerun })); } catch {}
-    const names = files.map((f) => "`" + f.name + "`").join(" or ");
+    try { writeFileSync(join(path, ".symbiot", WAITING), JSON.stringify({ step, files: files.map((f) => ({ ...f, sig: fileSig(f.path) })), cmd: runCmd(path), at: Date.now(), rerun: !!opts.rerun })); } catch {}
+    const on = waitOn(files.map((f) => f.name), settingsStep(step));
     out.yours = yours; if (files.length) out.waitFiles = files.map((f) => f.name);
-    out.note = "Answers saved. That step is still yours to do. " + (files.length ? `Your agent ${opts.rerun ? "starts by itself" : "can start"} once ${names} changes${opts.rerun ? " (while Symbiot runs)" : ""}.` : "Once it's done, click Start it now on this folder.");
+    out.note = "Answers saved. That step is still yours to do. " + (on ? `Your agent ${opts.rerun ? "starts by itself" : "can start"} once ${on} changes${opts.rerun ? " (while Symbiot runs)" : ""}.` : "Once it's done, click Start it now on this folder.");
     return out;
   }
   if (opts.rerun) { const e = runHandoff(path); if (e && e.busy) out.note = "Answers saved. An agent is still running in that folder, so another wasn't started. Send them again once it finishes."; else if (e && e.blocked) out.note = "Answers saved. " + e.note; else if (e) out.rerun = e.id; else out.note = "Answers saved. Set an agent command in Settings to have the agent pick them up automatically."; }
@@ -535,4 +557,4 @@ function agentsList() {
   }).concat(earlierRuns());
 }
 
-export { HANDOFFS, HANDOFF_PROMPT, QUESTIONS_MAX, shSingle, CLAUDE_CMD, ORCA_CLAUDE_CMD, handoffCmd, setHandoffCmd, grantAgent, grantRule, allowTool, claudeConnectors, withConnectors, connectorsInfo, fillHandoff, runHandoff, blockedAgain, runningHandoff, loadRuns, earlierRuns, namedFiles, waitingFor, startWaiting, writeTasks, releaseHeldTasks, startHeldTasks, detectHandoffs, orcaHandoffCmd, migrateOrcaCmd, migrateClaudeCmd, track, agentChanges, parseQuestions, suggestionTarget, agentQuestions, answerQuestions, agentsList };
+export { HANDOFFS, HANDOFF_PROMPT, QUESTIONS_MAX, OPTIONS_SHOWN, IDEAS_SHOWN, shSingle, CLAUDE_CMD, ORCA_CLAUDE_CMD, handoffCmd, setHandoffCmd, grantAgent, grantRule, allowTool, claudeConnectors, withConnectors, connectorsInfo, fillHandoff, runHandoff, blockedAgain, runningHandoff, loadRuns, earlierRuns, namedFiles, waitingFor, startWaiting, writeTasks, releaseHeldTasks, startHeldTasks, detectHandoffs, orcaHandoffCmd, migrateOrcaCmd, migrateClaudeCmd, track, agentChanges, parseQuestions, suggestionTarget, agentQuestions, answerQuestions, agentsList };

package/index.mjs

@@ -39,7 +39,7 @@ import { buildTasksMd, taskType, shipChanges, shipWithBump, bumpOffer, learnNpm,
 import { produce, mailState, setMail, sentMail } from "./writeups.mjs";
 import { loadScreens, screenImage, blueprint } from "./screens.mjs";
 import { mapPage, wholePage, pressRegion, typeRegion, scrollPage, signIn, isTrusted } from "./headless.mjs";
-import { watchState, addWatch, removeWatch, seenWatch, newsSince, markNews, checkWatch, setBrief, draftReply, watchBoard, boardLine } from "./watch.mjs";
+import { watchState, addWatch, removeWatch, seenWatch, newsSince, markNews, checkWatch, setBrief, draftReply, watchBoard, boardLine, boardChat, boardTalk, clearBoardChat } from "./watch.mjs";
 import { PORT as PHONE_PORT, phoneState, pairComputer, pollComputer, forgetComputer } from "./phone.mjs";
 import { startApp, updateCmd, isAppRunningWeekly } from "./server.mjs";
 
@@ -380,7 +380,7 @@ async function cmdScreens() {
 // ---- `symbiot watch`: pages Symbiot keeps track of, and what's new on them ----
 // new, board, seen, add, remove and check print JSON, for you or an agent.
 async function cmdWatch() {
-  const [sub = "list", a1] = argv.slice(1).filter((x, i, all) => !x.startsWith("--") && !["--every", "--hours"].includes(all[i - 1]));
+  const [sub = "list", a1, a2] = argv.slice(1).filter((x, i, all) => !x.startsWith("--") && !["--every", "--hours"].includes(all[i - 1]));
   const out = (x) => { console.log(JSON.stringify(x, null, 2)); if (x && x.error) process.exitCode = 1; };
   const hours = Number(flag("hours", 24)) || 24;
   if (sub === "list") {
@@ -404,6 +404,17 @@ async function cmdWatch() {
     const where = r.chat ? "types it into the chat's message box in Symbiot's browser, never sent" : "leaves it in Drafts, never sent";
     return out(r.ok ? { ...r, next: `Your agent is writing the reply and ${where}. Its log is in ` + join(r.dir, ".symbiot", "agent.log") + " (and the app's Agents tab)." } : r);
   }
+  // a card's chat, the same talk as 💬 on the Dashboard (through the app when it
+  // runs, so an agent it starts shows in its Agents tab): a reply drafted from that
+  // card brings it along. No question: the talk so far; --clear starts it over.
+  if (sub === "chat") {
+    if (!a1) return out({ error: "Give the watch's id: symbiot watch board lists them. Then  symbiot watch chat <id> \"what needs me?\"" });
+    if (has("clear")) return out((await viaApp("/api/watch/chat/clear", { id: a1 })) || clearBoardChat(a1));
+    if (!a2) return out(boardTalk(a1));
+    const r = (await viaApp("/api/watch/chat", { id: a1, question: a2 })) || await boardChat(a1, a2);
+    if (r.error === "not-connected") return out({ error: "Connect a model first: symbiot login (or Settings in the app)." });
+    const { chat, ...rest } = r; return out(chat ? { ...rest, messages: chat.length } : rest);
+  }
   if (sub === "add") { const screen = loadScreens().some((s) => s.id === a1) ? a1 : ""; return out(addWatch({ screen, site: screen ? "" : a1, every: flag("every", 15) })); }
   if (sub === "remove") return out(removeWatch(String(a1 || "")));
   if (sub === "brief") { if (a1 === "on" || a1 === "off") return out(setBrief(a1 === "on")); return out({ brief: watchState().brief, briefs: watchState().briefs }); }
@@ -428,6 +439,10 @@ async function cmdWatch() {
                                                empty line when nothing's new)
   symbiot watch seen <id>                      set a card's count back to 0, like its Seen
                                                button (what it found stays in watch new)
+  symbiot watch chat <id> "question"           talk over what's new on a card with your AI,
+                                               like 💬 on the Dashboard: a reply drafted from
+                                               that card follows what you agreed (JSON)
+  symbiot watch chat <id> [--clear]            the talk so far, or start it over
   symbiot watch check [id]                     read them now (JSON)
   symbiot watch draft <id>                     your agent drafts a reply to that new email
                                                ("mail": true in watch new) and leaves it in

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "symbiot",
-  "version": "0.43.0",
+  "version": "0.43.1",
   "description": "Your week, written from your real work. Reads your local git and writes it up with the AI of your choice — Claude, OpenAI, Gemini, or a local model.",
   "type": "module",
   "bin": {

package/tasks.mjs

@@ -7,7 +7,7 @@ import { join } from "node:path";
 import { readFileSync, writeFileSync, existsSync, readdirSync } from "node:fs";
 import { randomBytes } from "node:crypto";
 import { VERSION, LATEST_VERSION, REGISTRY, semverGt, loadConfig, saveConfig, loadTasks, saveTasks, taskWords, sameTask, uniqueTasks, sh, hasCmd, repoState } from "./core.mjs";
-import { QUESTIONS_MAX, handoffCmd, runningHandoff, writeTasks, startHeldTasks } from "./agents.mjs";
+import { QUESTIONS_MAX, OPTIONS_SHOWN, IDEAS_SHOWN, handoffCmd, runningHandoff, writeTasks, startHeldTasks } from "./agents.mjs";
 import { gitDefaultBranch } from "./drift.mjs";
 import { repoPathMap, openWork, detectRepo } from "./scan.mjs";
 
@@ -436,11 +436,12 @@ function buildTasksMd(name, ctx, list) {
   for (const ty of keys) { L.push(`### ${ty}`); for (const t of byType[ty]) L.push(`- [ ] ${t.text}`); L.push(""); }
   L.push("## When you finish an item", "- Tick it here (`- [x]`) as soon as it's done — that's how it reaches review. Ticking doesn't archive it: the user approves it in Symbiot, which commits it on a branch and opens a PR.", "- Leave your changes **uncommitted**, and don't tick anything you didn't finish or couldn't verify.", "");
   L.push("## If you need a decision, or have ideas", "You may be running unattended, so you can't ask in chat. Write `.symbiot/QUESTIONS.md` instead: Symbiot shows it to the user on your block in its Agents tab, and their answers come back in `.symbiot/ANSWERS.md` (read that first if it exists).",
-    `- At most ${QUESTIONS_MAX} questions, under a \`## Questions\` heading. Each is a \`### \` heading, then an optional line of context, then 2–4 options as \`- \` bullets (put the one you recommend first).`,
+    `- At most ${QUESTIONS_MAX} questions, under a \`## Questions\` heading. Each is a \`### \` heading, then a line of context, then exactly ${OPTIONS_SHOWN} options as \`- \` bullets, the one you recommend first, marked \`(recommended)\`. Symbiot shows only the first ${OPTIONS_SHOWN}; the user can always answer in their own words.`,
+    "- Judge the options before you ask. Most people pick the recommended option without weighing the other, and Symbiot works for a whole company (developers, sales, everyone), not one person, so the choice is really yours. Both options must be good routes to the best solution, never filler or one you wouldn't take. Each says in plain words, with no jargon, what it does and what it changes from then on for the project, the people working on it and the company. Recommend the one that's best for, in this order, the company, the people doing the work, then the task's goal. Base that on evidence you can check here (git history, tests, logs, how it's used, the answers so far), not on what's quickest, and give that evidence in the context line in a sentence.",
     "- Start every option with who acts, because the user reads it as something they'd be agreeing to do: `👤 You:` when they have to do something (a setting, a click, a command), `🤖 Agent:` when picking it is enough and the next run does the work. If it takes both, put their step first: `👤 You: allow npm install in Settings. 🤖 Agent: the next run adds ESLint`. Tag actions only: an option that just reports what the user saw or decides (`it notified me`, `not tried yet`, `keep it as it is`) gets no tag.",
     "- A `👤 You:` answer holds the next run back until the user's step is done. If their step changes a file, name the file in backticks (`.env`, `~/.termux/termux.properties`): the next run starts once it changes.",
     "- If a step needs a release that isn't out yet, name its version in the question (\"Once 0.41.0 is installed: …\"). Symbiot shows it next to the installed and npm versions, and holds back an answer that starts \"Done\" until that release is out and installed.",
-    "- Ideas, options or follow-ups outside these tasks go under `## Suggestions` as `- ` bullets — the user can add them to their tasks in one click.",
+    `- Ideas, options or follow-ups outside these tasks go under \`## Suggestions\` as \`- \` bullets. The user can add them to their tasks in one click, and often adds every one, so judge them the same way: only ideas you'd recommend, best first, each saying what it changes and why it's worth doing. Symbiot shows them ${IDEAS_SHOWN} at a time.`,
     "- An idea for a different project than this one (Symbiot itself, say, the app that sent you this brief) starts with that project's folder name, `- [repo: symbiot] …`, so it goes to that project's tasks instead of this one's.",
     "- Carry on with everything that doesn't depend on an answer, and don't tick an item that does. Ask there rather than doing anything destructive.", "");
   L.push("---", 'To action these, tell your coding agent: "Read `.symbiot/TASKS.md` and implement the unchecked items in this repo, using the context above. Tick each item as you finish it and leave changes uncommitted for review. Confirm with me before anything destructive."');

package/ui.mjs

@@ -537,8 +537,10 @@ qs.forEach(function(q,i){var rl=q.release;h+="<div class='q' data-i='"+i+"'><div
 (q.options||[]).forEach(function(o,j){var off=rl&&rl.waiting&&DONEOPT.test(o);h+="<label class='opt"+(off?" off":"")+"'><input type='radio' name='q_"+esc(a.id)+"_"+i+"' value='"+j+"'"+(off?" disabled":"")+"><span>"+whoHtml(o)+(off?" <i>(once "+esc(rl.name+" "+rl.needs)+" is installed)</i>":"")+"</span></label>";});
 h+="<input class='qother' placeholder='"+((q.options&&q.options.length)?"or answer in your own words":"your answer")+"'></div>";});
 h+="<div class='row'><button class='act qsend' title='save the answers and hand the repo back to your agent'>Send answers &amp; continue</button><button class='ghost qsave' title='save the answers for the next run'>Save only</button></div>";}
-if(ss.length){h+="<h4>&#128161; Ideas from the agent</h4>";ss.forEach(function(s,i){h+="<div class='idea'><span style='flex:1'>"+esc(s.text)+(s.other?" <span class='tag' title='this idea is for another project, so + task adds it to that one'>for "+esc(s.repo)+"</span>":"")+"</span>"+(s.added?"<span class='tag'>in Tasks</span>":"<button class='ghost qidea' data-i='"+i+"' style='padding:3px 9px;font-size:12px'>+ task</button>")+"</div>";});}
+if(ss.length){var shown=IDEASOPEN[a.path]?ss.length:(k.ideasShown||ss.length);h+="<h4>&#128161; Ideas from the agent</h4>";ss.forEach(function(s,i){if(i>=shown)return;h+="<div class='idea'><span style='flex:1'>"+esc(s.text)+(s.other?" <span class='tag' title='this idea is for another project, so + task adds it to that one'>for "+esc(s.repo)+"</span>":"")+"</span>"+(s.added?"<span class='tag'>in Tasks</span>":"<button class='ghost qidea' data-i='"+i+"' style='padding:3px 9px;font-size:12px'>+ task</button>")+"</div>";});
+if(ss.length>shown)h+="<div class='row'><button class='ghost qmore' title='the agent ranks its ideas best first: weigh these two before adding more'>"+(ss.length-shown)+" more idea"+(ss.length-shown>1?"s":"")+", ranked lower</button></div>";}
 return h+"</div>";}
+var IDEASOPEN={};
 function qKeyOf(qe){var box=qe.closest('.aq');var a=box&&agentById(box.getAttribute('data-id'));var q=a&&a.ask.questions[+qe.getAttribute('data-i')];return q?a.path+'|'+q.q:'';}
 function saveDrafts(el){el.querySelectorAll('.aq .q').forEach(function(qe){var k=qKeyOf(qe);if(!k)return;var pick=qe.querySelector('input[type=radio]:checked');QDRAFT[k]={o:pick?pick.value:'',t:qe.querySelector('.qother').value};});}
 function restoreDrafts(el){el.querySelectorAll('.aq .q').forEach(function(qe){var d=QDRAFT[qKeyOf(qe)];if(!d)return;qe.querySelectorAll('input[type=radio]').forEach(function(r){r.checked=!r.disabled&&r.value===d.o;});qe.querySelector('.qother').value=d.t||'';});}
@@ -553,10 +555,11 @@ if(!r||r.error){box.querySelectorAll('button').forEach(function(b){b.disabled=fa
 msg.innerHTML="<div class='note ok'>&#10003; Saved "+r.saved+" answer"+(r.saved>1?"s":"")+" for <b>"+esc(a.name)+"</b> in .symbiot/ANSWERS.md"+(r.rerun?" &middot; your agent is picking them up now.":".")+(r.yours&&r.yours.length?"<div class='err'>&#128100; Still yours to do: "+esc(r.yours.join(' '))+"</div>":"")+(r.note?"<div class='muted'>"+esc(r.note)+"</div>":"")+"</div>";loadAgents();});}
 var s1=box.querySelector('.qsend'),s2=box.querySelector('.qsave');
 if(s1)s1.addEventListener('click',function(){send(true);});if(s2)s2.addEventListener('click',function(){send(false);});
+var more=box.querySelector('.qmore');if(more)more.addEventListener('click',function(){IDEASOPEN[a.path]=true;loadAgents();});
 box.querySelectorAll('.qidea').forEach(function(btn){btn.addEventListener('click',function(){var s=a.ask.suggestions[+btn.getAttribute('data-i')];if(!s)return;btn.disabled=true;
 api('/api/tasks/add',{text:s.text,repo:s.repo||a.name}).then(function(){s.added=true;btn.outerHTML="<span class='tag'>in Tasks</span>";});});});});}
 // A step of yours an answer picked (👤 You): the next run waits for it (agents.mjs waitingFor)
-function waitHtml(a){var w=a.waiting;if(!w)return '';var fs=(w.files||[]).map(function(f){return "<code>"+esc(f)+"</code>";}).join(" or ");
+function waitHtml(a){var w=a.waiting;if(!w)return '';var fs=(w.files||[]).map(function(f){return "<code>"+esc(f)+"</code>";}).join(" or ");if(w.cmd)fs=(fs?fs+" or ":"")+"the agent command in Settings &rarr; Handoff";
 return "<div class='aq'><h4>&#9208; Waiting on your step</h4><div class='qt'>&#128100; "+esc(w.step)+"</div><div class='qc'>"+(fs?"Your agent "+(w.rerun?"starts by itself":"can start")+" once "+fs+" changes"+(w.rerun?", while Symbiot runs":"")+". Done it some other way?":"Your agent waits for it, so it doesn't stop on the same questions again.")+" Start it now once it's done.</div><div class='row'><button class='act wstart' data-id='"+esc(a.id)+"'>Start it now</button></div></div>";}
 function wireWaits(el){el.querySelectorAll('.wstart').forEach(function(btn){btn.addEventListener('click',function(){var a=agentById(btn.getAttribute('data-id'));if(!a)return;btn.disabled=true;
 api('/api/open',{path:a.path,force:true}).then(function(x){var msg=document.getElementById('agentsmsg');msg.innerHTML=x&&x.opened?"<div class='note ok'>&#10003; Started your agent in <b>"+esc(a.name)+"</b>.</div>":"<div class='note err'>"+esc(a.name)+": "+(x&&x.busy?"an agent is already running there.":"it didn&#39;t start. Check the command in Settings.")+"</div>";loadAgents();}).catch(function(e){btn.disabled=false;document.getElementById('agentsmsg').innerHTML="<div class='note err'>"+esc(String((e&&e.message)||e))+"</div>";});});});}

package/watch.mjs

@@ -280,6 +280,8 @@ async function boardChat(id, question, { hours = 72, now = Date.now(), ask = wri
   saveWatch(d2);
   return { answer, chat: w2.chat, ...(peeks.length ? { links: peeks } : {}), ...(r.did ? { did: r.did } : {}) };
 }
+// The card's talk so far (symbiot watch chat <id>), [{ role, text, ts }].
+function boardTalk(id) { const w = loadWatch().watches.find((x) => x.id === id); return w ? { chat: w.chat || [] } : { error: `No watch ${id}.` }; }
 function clearBoardChat(id) { const d = loadWatch(), w = d.watches.find((x) => x.id === id); if (!w) return { error: `No watch ${id}.` }; delete w.chat; saveWatch(d); return { ok: true }; }
 // The card's talk, for a draft's brief: "" when there's been none.
 function talkOf(w) {
@@ -493,4 +495,4 @@ function startWatches(opts = {}) {
   return () => { clearTimeout(first); clearInterval(every); };
 }
 
-export { LINK_ASK, WATCH_FILE, EVERY, GITHUB_INBOX, DRAFTS_DIR, itemsOf, itemKey, newItems, remember, isGitHubInbox, githubItems, readGitHub, setBrief, briefOf, newsNotice, markNews, watchState, addWatch, setEvery, removeWatch, clearNews, seenWatch, newsSince, newsAfter, waitingOn, watchBoard, boardLine, boardChat, clearBoardChat, talkOf, isMail, isChat, draftsUrl, draftBrief, chatBrief, draftReply, openChat, checkWatch, dueWatches, startWatches };
+export { LINK_ASK, WATCH_FILE, EVERY, GITHUB_INBOX, DRAFTS_DIR, itemsOf, itemKey, newItems, remember, isGitHubInbox, githubItems, readGitHub, setBrief, briefOf, newsNotice, markNews, watchState, addWatch, setEvery, removeWatch, clearNews, seenWatch, newsSince, newsAfter, waitingOn, watchBoard, boardLine, boardChat, boardTalk, clearBoardChat, talkOf, isMail, isChat, draftsUrl, draftBrief, chatBrief, draftReply, openChat, checkWatch, dueWatches, startWatches };

package/writeups.mjs

@@ -69,6 +69,12 @@ function extractJson(s) {
   if (a >= 0 && b > a) { try { return JSON.parse(t.slice(a, b + 1)); } catch {} }
   return null;
 }
+// Two at a time, judged (as agents' questions are, agents.mjs OPTIONS_SHOWN):
+// most people act on whatever's suggested without weighing it, so the model
+// picks, and says why in a line.
+const JUDGE = `Judge them before you write them: people tend to do whatever is listed without weighing it, and this is for a whole company, not one person. ` +
+  `List only what you'd recommend, best first, and choose by what's best for, in this order, the company, the people doing the work, then the project's goal. ` +
+  `For each, say in plain words (no jargon) what it changes from then on and the evidence for it. `;
 // AI "value" suggestions for one repo, from its recent commits + open work.
 async function repoSuggest(path) {
   if (!resolveProvider()) return { error: "not-connected" };
@@ -81,8 +87,8 @@ async function repoSuggest(path) {
   const footer = reportFooter(path, auth, state, rd);
   const system =
     `You are a pragmatic senior engineer advising on one project. Write exactly three short sections with headings: ` +
-    `"In flight" (what's clearly underway), "Next steps" (3-5 concrete actions), "Ideas" (2-3 that fit where this is heading). ` +
-    `Ground each point in the evidence; recent commits beat the README. Do not advise against the stated conventions. ` +
+    `"In flight" (what's clearly underway), "Next steps" (the 2 best concrete actions), "Ideas" (the 2 best that fit where this is heading). ` +
+    JUDGE + `Recent commits beat the README. Do not advise against the stated conventions. ` +
     (state.stale ? `CRITICAL: the working tree is a STALE checkout — never suggest committing it (would revert history). ` : ``) +
     `No filler, no preamble.`;
   const prompt =
@@ -107,8 +113,8 @@ async function folderSuggest(path) {
   const system =
     `You are a pragmatic senior engineer looking at a project FOLDER that is NOT under version control. ` +
     `From its files, write exactly three short sections with headings: "Overview" (what this project is, 1-2 sentences), ` +
-    `"Next steps" (3-5 concrete actions — a strong first one is often "git init" if this looks like real work), ` +
-    `"Ideas" (2-3 that fit where it's heading). Ground every point in the actual files/manifest/README shown. No preamble.`;
+    `"Next steps" (the 2 best concrete actions — a strong first one is often "git init" if this looks like real work), ` +
+    `"Ideas" (the 2 best that fit where it's heading). ` + JUDGE + `The evidence here is the files, manifest and README shown. No preamble.`;
   const prompt =
     `Folder: ${det.name}\nStack: ${stack}\nFiles (${det.files}):\n${rels.slice(0, 120).join("\n")}\n\n` +
     (manifest ? `Manifest:\n${manifest}\n\n` : "") + (readme ? `README excerpt:\n${readme}\n\n` : "") + `Write the overview and suggestions.`;