mixdog 1.2.5 → 1.3.0

package/src/defaults/skills/setup-guide/SKILL.md

@@ -0,0 +1,175 @@
+---
+name: setup-guide
+description: Guide a new user through first setup and explain Mixdog concepts.
+when_to_use: 'Mixy''s first chat; how to start, what to do first, what a Mixdog term means; not one named setting.'
+dependencies:
+  tools:
+    - type: tool
+      value: setup
+---
+
+# Setup guide
+
+Walk a new user through Mixdog's first setup one step at a time, and explain
+Mixdog's concepts in plain words. Changes go through the `setup` tool, never
+through file edits. One named setting the user already chose belongs to `setup`.
+
+## Style
+
+Reply as in agent mode: one step or one concept per message, short plain
+sentences, no tables or headings, in the user's language. In Korean use the
+formal polite style (~합니다 / ~하세요, never ~해요). Say "봇" for Mixy and
+other contacts, "에이전트" only for workflow agents. Never print ids, file
+names or dotted internal names. End a step with one question, not a list of options.
+
+## Mode
+
+- Setup walk-through: the user accepted setup help ("기본 설정부터 같이 할까요?"),
+  or asks how to start or what to do first.
+- Concept question: answer it from the glossary below, then stop. Go deeper
+  only when asked, and offer the matching setup step only if it is not done.
+
+## Opening
+
+Start in this order, one message each:
+
+1. Say you will go through setup one thing at a time.
+2. Check `status features` and `status mcp` first. Offer add-ons tier by tier;
+   finish a tier before the next. Never mention a feature that is already on
+   or installed, and skip a tier with nothing missing.
+3. Then the connectors, then the steps below.
+
+Tier 1, essential: Media generation, Git & GitHub (`git`), Visualize. If any
+is missing or off, say it is essential and ask once for all of them
+("필수 기능인데 설치할까요?").
+
+Tier 2, recommended: Memory (`memory`), Computer Use (`computer`), Browser Use
+(`browser`). Offer the missing ones in one short line.
+
+Tier 3, optional: Voice transcription (`voice`), Local Provider
+(`localProvider`), Code Tidy (`tidy`). Offer the missing ones in one short line.
+
+Install only what the user accepts. How each is done:
+
+- `install_builtin` with the name for git, memory, autoEffort, computer,
+  browser, voice, localProvider and tidy. Git installs system Git through the
+  Desktop when this chat is open there. Auto reasoning can be unsupported on
+  a machine: report that instead of retrying. Computer Use and Browser Use
+  need the Desktop, and the user approves OS permission dialogs. Local
+  Provider only prepares the runtime; models belong to `local-provider`.
+- Memory is install-only; its later on/off is `set_memory_enabled`.
+- Visualize has no install: `set_builtin_enabled` turns it on.
+- Media generation has no setup action: `open plugins` and let the user turn
+  it on there, then re-check `status features`.
+
+Connectors come after the tiers. First explain in a sentence or two: a
+connector links an outside service so bots and agents can read and act in it,
+and anything that sends or posts asks first through an approval card. Then
+suggest by what the user does, only these: Gmail (with a Google app password;
+it needs 2-Step Verification on the Google account, and the connect window
+opens the app-password page in the browser), GitHub (comes with Git & GitHub),
+Notion, Linear or Jira. Do not suggest Google Calendar or Google
+Drive; their Google sign-in is still under review. Connect one with
+`connect_connector`; it only leaves a card, so tell the user to press it and
+confirm with `status mcp`.
+
+## Walk-through
+
+The steps mirror the desktop starter guide and its checklist; do not keep
+separate progress. Read state first with `setup` `status` (`providers`,
+`onboarding`, `mcp`, `plugins`, `agents`, `connection`) and skip steps already
+done. Take the first open step, explain it in a sentence or two, ask, then act.
+
+1. provider: done when a provider is authenticated (`status providers`). Sign-in
+   and API keys are entered by the user: `open providers`, never ask for a key
+   in chat. Suggest a subscription sign-in (Claude, ChatGPT, Gemini and the
+   like) first; an API key only when the user has no subscription or asks.
+2. project: done when a Project exists or `onboarding` lists `project` as
+   visited. A folder is optional. Register one with `save_project` for a path
+   the user names.
+3. workflow: done when `onboarding` lists `workflow` as visited. Explain
+   workflows and agents, `open workflows` so the user can turn agents on and
+   pick models (`set_agent_route` once they name a model). Recommend models
+   from `${MIXDOG_SKILL_DIR}/model-recommendations.json`, role by role, only
+   among providers the user is signed in to (`status providers`) and only a
+   model the provider's model list actually offers; picks are an ordered list
+   of any length, so recommend the first one that qualifies and otherwise take
+   the next. A subscription sign-in always comes first: when the user has one,
+   take the first qualifying pick it offers before any pick that would bill an
+   API key, and only fall back to API-key providers when no subscription
+   offers one. If `asOf` is older than `staleAfterDays`, say the list may be out of
+   date. Never print ids or file names. Roles with
+   `preferOtherProviderThan` (Reviewer, Advisor) work best with the opposite
+   tendency from the model doing the work: pick the first qualifying model whose
+   provider differs from the chosen Lead's and Heavy Worker's providers, and
+   fall back to the plain first pick only when the user has a single signed-in
+   provider. Give the reason in one line: a different model family catches what
+   the main one misses. Suggest turning on the Maintainer with
+   a cheap model at the lowest effort; it also decides turns in group rooms.
+   A pick may carry `plans`: when the Lead provider's plan (`plan` in
+   `status providers`) is listed there, recommend that model instead. A null or
+   unlisted plan keeps the pick as is; never fail on an unknown plan.
+   Delegation and roster come from `delegation` in the data file. First ask
+   once how the user likes to work: one agent doing everything, or splitting
+   work across agents. Say plainly that Solo is a good choice at any plan: it
+   is the most efficient, and with a strong model its quality is often not much
+   lower; delegation trades more usage for speed and parallel work. Then suggest
+   the mode whose `plans` lists the Lead provider's plan (the default
+   suggestion), but the user's preference overrides it. Ask once; set it with
+   `set_orchestration_mode`, and say it can be changed later in settings. The
+   roster follows the chosen mode: turn on only `baseRoster` plus that mode's
+   `agents`, using `set_agent_route` with `route.disabled` set false for what
+   the user accepts and true for the rest; do not enable others. Offer
+   `leanReview` for Solo or focused as a cost saver: the Lead reviews its own
+   work, Reviewer stays off, Advisor is on but used rarely (only when stuck).
+   Advisor in balanced or swarm is a light, occasional second opinion from a
+   different model family at key decision points, not on every task. Say in
+   one line that more agents can be turned on later on the Workflows page.
+   Right after the models are picked, explain Auto reasoning: it adjusts
+   reasoning depth per message, so hard work may come out slightly lower in
+   quality but cost is much lower. Ask whether to turn it on
+   (`install_builtin` `autoEffort`). If declined, say in one line that it can
+   be turned on later from the model picker.
+4. remote: done when a device is paired (`status connection`) or `remote` is
+   visited. Pairing is user-operated: `open connection`.
+5. extensions: done when `extensions` is visited. Explain plugins and skills;
+   offer only built-ins not already covered by the tiers above.
+6. mixy: done once Mixy exists, which is already true when this skill runs in
+   Mixy's chat. Offer to make another bot when the user wants one.
+
+After a change, report it in a sentence and move on. The user may stop at any
+step; do not insist.
+
+## Glossary
+
+- Bot (봇): an agent-mode contact you chat with. Mixy is the representative
+  bot; the user can have it make others, such as a GitHub Watcher.
+- Agent (에이전트): a workflow sub-agent such as Worker or Reviewer that
+  does part of a task. Users add their own on the Workflows page with the
+  "에이전트 +" button. Bots are people-facing; agents work behind a task.
+- Default agents (기본 에이전트): built-in helpers listed on Workflows, such as
+  Web Search (runs the web search tool) and the Maintainer (memory upkeep and
+  session titles). They need a model route to work.
+- Sessions and projects: a coding session runs in a Project, a folder whose
+  instructions and memories it loads.
+- Workflows and work rules: a workflow sets how a request gets done; work
+  rules are the instructions it applies.
+- Connectors and extensions: connectors are MCP services such as GitHub or
+  Notion (`connect_connector`, the user presses the card); extensions are
+  plugins that add tools and skills that teach how to work.
+- Group rooms (그룹 대화방): one chat with several bots. `@name` or "봇 부르기"
+  addresses one bot; otherwise the Maintainer decides who answers next.
+- Schedules and webhooks: a timed or incoming-event trigger (`save_automation`).
+  Each run is one turn in its bot's own chat, with that bot's persona, rules,
+  permissions, memory and model. Without a bot they cannot be used.
+- Approval cards: when a tool needs permission it waits in the chat until the
+  user approves or declines.
+- Phone remote access: pair a phone under Settings → Connection to follow and
+  steer tasks away from the PC.
+
+## Pitfalls
+
+- Do not claim a step is done from memory; re-read state after acting.
+- A UI handoff (`open`) is not completion; ask the user to confirm, then
+  re-check the status.
+- Do not run several steps in one message or queue the next question early.

package/src/defaults/skills/setup-guide/model-recommendations.json

@@ -0,0 +1,333 @@
+{
+  "asOf": "2026-10-20",
+  "staleAfterDays": 14,
+  "delegation": {
+    "note": "Default suggestion by the Lead provider's plan (status providers, plan). The user's working style overrides it; the roster follows the chosen mode. Solo is a valid choice at any plan.",
+    "unknownPlan": "ask the user which plan they have; if they do not know, suggest focused",
+    "baseRoster": ["Maintainer", "Web Search"],
+    "leanReview": {
+      "fitsModes": ["none", "focused"],
+      "description": "Cost-saving alternative to a Reviewer agent: the Lead reviews its own work, and the Advisor is called only when work gets stuck (errors recur, results stop converging).",
+      "agentsOff": ["Reviewer"],
+      "agentsOn": ["Advisor"],
+      "advisorUse": "only when stuck; rarely"
+    },
+    "modes": [
+      {
+        "mode": "none",
+        "label": "Solo",
+        "agents": [],
+        "advisorUse": null,
+        "plans": {
+          "Anthropic": ["none"],
+          "OpenAI": ["free", "none"],
+          "Google": ["free-tier"],
+          "other": "API key only or no subscription"
+        }
+      },
+      {
+        "mode": "focused",
+        "label": "Focused",
+        "agents": ["Worker", "Reviewer"],
+        "advisorUse": null,
+        "plans": { "Anthropic": ["pro"], "OpenAI": ["plus"], "Google": ["standard-tier", "legacy-tier"] }
+      },
+      {
+        "mode": "balanced",
+        "label": "Balanced",
+        "agents": ["Worker", "Reviewer", "Heavy Worker", "Advisor"],
+        "advisorUse": "sparingly: a light second opinion from a different model family only at key decision points (before choosing an approach, when results stop converging), not on every task; it runs rarely so it costs little",
+        "plans": {
+          "Anthropic": ["team", "enterprise"],
+          "OpenAI": ["team", "business", "enterprise"],
+          "Google": ["g1-pro-tier"]
+        }
+      },
+      {
+        "mode": "swarm",
+        "label": "Swarm",
+        "agents": ["Worker", "Reviewer", "Heavy Worker", "Advisor", "Writer", "Front Worker"],
+        "advisorUse": "sparingly, same as balanced",
+        "plans": { "Anthropic": ["max"], "OpenAI": ["pro"], "Google": ["g1-ultra-tier"] }
+      }
+    ]
+  },
+  "roles": [
+    {
+      "roles": ["Lead", "Bot default"],
+      "picks": [
+        {
+          "provider": "Anthropic",
+          "model": "Claude Opus 5.5",
+          "effort": "medium",
+          "plans": { "pro": "Claude Sonnet 5.5" },
+          "checkedAt": "2026-10-20",
+          "source": "https://benchlm.ai/anthropic/api-pricing"
+        },
+        {
+          "provider": "OpenAI",
+          "model": "GPT-6.1 Sol",
+          "effort": "medium",
+          "checkedAt": "2026-10-20",
+          "source": "https://benchlm.ai/openai/api-pricing"
+        },
+        {
+          "provider": "DeepSeek",
+          "model": "DeepSeek V4.1 Flash",
+          "effort": "none",
+          "note": "budget; no effort setting",
+          "checkedAt": "2026-10-20",
+          "source": "https://api-docs.deepseek.com/quick_start/pricing/"
+        }
+      ]
+    },
+    {
+      "roles": ["Heavy Worker"],
+      "picks": [
+        {
+          "provider": "Anthropic",
+          "model": "Claude Opus 5.5",
+          "effort": "medium",
+          "plans": { "pro": "Claude Sonnet 5.5" },
+          "checkedAt": "2026-10-20",
+          "source": "https://platform.claude.com/docs/en/build-with-claude/effort"
+        },
+        {
+          "provider": "OpenAI",
+          "model": "GPT-6.1 Sol",
+          "effort": "high",
+          "note": "xhigh for the hardest long runs",
+          "checkedAt": "2026-10-20",
+          "source": "https://openai.com/index/practical-guide-building-gpt-6/"
+        },
+        {
+          "provider": "Google",
+          "model": "Gemini 4 Argon",
+          "effort": "high",
+          "note": "budget",
+          "checkedAt": "2026-10-20",
+          "source": "https://ai.google.dev/gemini-api/docs/generate-content/latest-model"
+        }
+      ]
+    },
+    {
+      "roles": ["Advisor"],
+      "preferOtherProviderThan": ["Lead", "Heavy Worker"],
+      "picks": [
+        {
+          "provider": "OpenAI",
+          "model": "GPT-6 Astra",
+          "effort": "xhigh",
+          "checkedAt": "2026-10-20",
+          "source": "https://openai.com/index/practical-guide-building-gpt-6/"
+        },
+        {
+          "provider": "Anthropic",
+          "model": "Claude Opus 5.5",
+          "effort": "xhigh",
+          "checkedAt": "2026-10-20",
+          "source": "https://platform.claude.com/docs/en/build-with-claude/effort"
+        },
+        {
+          "provider": "Google",
+          "model": "Gemini 4 Argon",
+          "effort": "high",
+          "note": "budget; high is its top level",
+          "checkedAt": "2026-10-20",
+          "source": "https://ai.google.dev/gemini-api/docs/generate-content/latest-model"
+        }
+      ]
+    },
+    {
+      "roles": ["Reviewer"],
+      "preferOtherProviderThan": ["Lead", "Heavy Worker"],
+      "picks": [
+        {
+          "provider": "OpenAI",
+          "model": "GPT-6 Astra",
+          "effort": "high",
+          "checkedAt": "2026-10-20",
+          "source": "https://www.benchleader.com/best/llm-for-coding"
+        },
+        {
+          "provider": "Anthropic",
+          "model": "Claude Opus 5.5",
+          "effort": "high",
+          "checkedAt": "2026-10-20",
+          "source": "https://benchlm.ai/anthropic/api-pricing"
+        },
+        {
+          "provider": "Google",
+          "model": "Gemini 4 Argon",
+          "effort": "high",
+          "note": "budget",
+          "checkedAt": "2026-10-20",
+          "source": "https://benchlm.ai/google/api-pricing"
+        }
+      ]
+    },
+    {
+      "roles": ["Writer"],
+      "picks": [
+        {
+          "provider": "Google",
+          "model": "Gemini 4 Argon",
+          "effort": "medium",
+          "checkedAt": "2026-10-20",
+          "source": "https://ai.google.dev/gemini-api/docs/generate-content/latest-model"
+        },
+        {
+          "provider": "Google",
+          "model": "Gemini 3.8 Flash",
+          "effort": "medium",
+          "checkedAt": "2026-10-20",
+          "source": "https://ai.google.dev/gemini-api/docs/generate-content/latest-model"
+        },
+        {
+          "provider": "Anthropic",
+          "model": "Claude Sonnet 5.5",
+          "effort": "medium",
+          "checkedAt": "2026-10-20",
+          "source": "https://benchlm.ai/anthropic/api-pricing"
+        },
+        {
+          "provider": "OpenAI",
+          "model": "GPT-6.1 Sol",
+          "effort": "medium",
+          "checkedAt": "2026-10-20",
+          "source": "https://benchlm.ai/openai/api-pricing"
+        }
+      ]
+    },
+    {
+      "roles": ["Worker"],
+      "picks": [
+        {
+          "provider": "Anthropic",
+          "model": "Claude Sonnet 5.5",
+          "effort": "medium",
+          "checkedAt": "2026-10-20",
+          "source": "https://benchlm.ai/anthropic/api-pricing"
+        },
+        {
+          "provider": "OpenAI",
+          "model": "GPT-6.1 Sol",
+          "effort": "medium",
+          "checkedAt": "2026-10-20",
+          "source": "https://benchlm.ai/openai/api-pricing"
+        },
+        {
+          "provider": "DeepSeek",
+          "model": "DeepSeek V4.1 Flash",
+          "effort": "none",
+          "note": "budget; no effort setting",
+          "checkedAt": "2026-10-20",
+          "source": "https://api-docs.deepseek.com/quick_start/pricing/"
+        }
+      ]
+    },
+    {
+      "roles": ["Front Worker"],
+      "picks": [
+        {
+          "provider": "Anthropic",
+          "model": "Claude Opus 5.5",
+          "effort": "medium",
+          "checkedAt": "2026-10-20",
+          "source": "https://www.benchleader.com/best/llm-for-coding"
+        },
+        {
+          "provider": "DeepSeek",
+          "model": "DeepSeek V4.1 Flash",
+          "effort": "none",
+          "note": "budget; no effort setting",
+          "checkedAt": "2026-10-20",
+          "source": "https://api-docs.deepseek.com/quick_start/pricing/"
+        },
+        {
+          "provider": "Google",
+          "model": "Gemini 4 Argon",
+          "effort": "medium",
+          "checkedAt": "2026-10-20",
+          "source": "https://ai.google.dev/gemini-api/docs/generate-content/latest-model"
+        },
+        {
+          "provider": "OpenAI",
+          "model": "GPT-6.1 Sol",
+          "effort": "medium",
+          "checkedAt": "2026-10-20",
+          "source": "https://benchlm.ai/openai/api-pricing"
+        }
+      ]
+    },
+    {
+      "roles": ["Web Search"],
+      "picks": [
+        {
+          "provider": "xAI",
+          "model": "Grok 4.7",
+          "effort": "none",
+          "checkedAt": "2026-10-20",
+          "source": "https://benchlm.ai/providers/xai"
+        },
+        {
+          "provider": "OpenAI",
+          "model": "GPT-6.1 Sol",
+          "effort": "low",
+          "checkedAt": "2026-10-20",
+          "source": "https://benchlm.ai/openai/api-pricing"
+        },
+        {
+          "provider": "Google",
+          "model": "Gemini 3.8 Flash",
+          "effort": "low",
+          "checkedAt": "2026-10-20",
+          "source": "https://benchlm.ai/google/api-pricing"
+        },
+        {
+          "provider": "Anthropic",
+          "model": "Claude Sonnet 5.5",
+          "effort": "low",
+          "checkedAt": "2026-10-20",
+          "source": "https://benchlm.ai/anthropic/api-pricing"
+        }
+      ]
+    },
+    {
+      "roles": ["Maintainer"],
+      "note": "The Maintainer also decides turns in group rooms; there is no separate role for that.",
+      "picks": [
+        {
+          "provider": "OpenAI",
+          "model": "GPT-6 Luna",
+          "effort": "low",
+          "checkedAt": "2026-10-20",
+          "source": "https://benchlm.ai/openai/api-pricing"
+        },
+        {
+          "provider": "DeepSeek",
+          "model": "DeepSeek V4.1 Flash",
+          "effort": "none",
+          "note": "no effort setting",
+          "checkedAt": "2026-10-20",
+          "source": "https://api-docs.deepseek.com/quick_start/pricing/"
+        },
+        {
+          "provider": "Anthropic",
+          "model": "Claude Haiku 5.5",
+          "effort": "low",
+          "checkedAt": "2026-10-20",
+          "source": "https://benchlm.ai/anthropic/api-pricing"
+        },
+        {
+          "provider": "Zhipu",
+          "model": "GLM-5.3-Flash",
+          "effort": "none",
+          "note": "budget",
+          "checkedAt": "2026-10-20",
+          "source": "https://www.benchleader.com/best/llm-for-coding"
+        }
+      ]
+    }
+  ]
+}

package/src/defaults/skills/visualize/SKILL.md

@@ -24,8 +24,9 @@ tradeoffs, and wording decisions stay in chat even when the topic is UI.
 - *Show* — any artifact you produced or changed (HTML, deck, document, sheet,
   PDF, image, video, a local dev page): `open` it once when it is ready. Edits
   reload the tab by themselves; never reopen to refresh.
-- *Mockup* — a UI change before implementation, or the user asks for a
-  시안/mockup/preview. §2.
+- *Mockup* — the user asks for a mockup or preview, or a UI change whose
+  look the user still has to decide. A change whose look is already settled
+  is built directly. §2.
 - *Options* — the direction is genuinely open (the user is unsure, or two
   readings of the request lead to different layouts). §3.
 - *Explain* — a flow, state machine, architecture, or data path the user needs

package/src/lib/rules-builder.cjs

@@ -139,7 +139,7 @@ function buildLanguageSection(dataDir) {
   // mistranslate, so the two clauses would be dead text.
   if (language.prompt !== 'English') {
     lines.push(
-      `- Preserve identifiers, paths, commands, API names, and exact errors. For technical jargon, use the established term or a plain functional description, never an invented literal translation.`
+      `- Preserve identifiers, paths, commands, API names, and exact errors. Keep technical terms in their established form, usually the original English (blocker, cache, commit); add a short plain explanation only when the reader may not know it. Never invent a literal translation.`
     );
   }
   return `# Language\n\n${lines.join('\n')}`;
@@ -361,17 +361,64 @@ function buildRouteRulesContent({
   return omitToolRoutes(parts.join('\n'), omitTools, allowTools);
 }
 
-function buildLeadRoleContent({ PLUGIN_ROOT, includeLeadBrief = true }) {
+// LEAD.md is split for BP3 prefix sharing (every Lead shares `common`; the
+// mode-variant passages and the identity line move later):
+// - `common`: every line but the identity line and the two passages below;
+// - `coding` / `agent`: those two passages as bullets — LEAD.md's own text
+//   for coding mode, the conversational variants for agent mode;
+// - `identity`: the coding identity sentence ("You are Mixdog, …").
+const LEAD_IDENTITY_LINE = /^- (You are Mixdog, a coding agent\.)\n/m;
+const LEAD_PROGRESS_PASSAGE = / (Before your first tool call, briefly state[\s\S]*?narration or plan repeats\.)/;
+const LEAD_REPORT_PASSAGE = / (On an agent report, review its diff[\s\S]*?not a completed request\.)/;
+const AGENT_MODE_PROGRESS =
+  '- Just answer conversation. Quick tool use (checking memory, leaving a\n' +
+  '  reaction) needs no announcement. When starting longer work (research,\n' +
+  '  delegation, multi-step tasks), say what you are doing, and share the\n' +
+  '  result when it finishes.';
+const AGENT_MODE_REPORT =
+  '- On an agent report, check the result against what you delegated and\n' +
+  '  relay it in your own words, keeping uncertainty, failures and sources\n' +
+  '  intact. Review a diff only when you delegated code.';
+
+function buildLeadRoleContent({ PLUGIN_ROOT, includeLeadBrief = true, part = 'common' }) {
   const RULES_DIR = path.join(PLUGIN_ROOT, 'rules');
-  const lead = readOptional(path.join(RULES_DIR, 'lead', 'LEAD.md'));
-  if (!lead) return '';
+  const raw = readOptional(path.join(RULES_DIR, 'lead', 'LEAD.md'));
+  if (!raw) return '';
   // Delegation-free sessions (headless, no delegatable agents) never expose the `agent` tool,
   // so Lead guidance marked `<!-- tools: agent -->` (briefing, completion
   // notifications) is dropped there and kept verbatim for delegating workflows.
-  return omitToolRoutes(lead, includeLeadBrief ? [] : ['agent']);
+  const lead = omitToolRoutes(raw, includeLeadBrief ? [] : ['agent']);
+  if (part === 'identity') return lead.match(LEAD_IDENTITY_LINE)?.[1] || '';
+  const progress = lead.match(LEAD_PROGRESS_PASSAGE);
+  const report = lead.match(LEAD_REPORT_PASSAGE);
+  if (part === 'coding' || part === 'agent') {
+    const agentMode = part === 'agent';
+    const bullets = [
+      progress && (agentMode ? AGENT_MODE_PROGRESS : `- ${progress[1]}`),
+      report && (agentMode ? AGENT_MODE_REPORT : `- ${report[1]}`),
+    ].filter(Boolean);
+    return bullets.length ? `# Progress and Reports\n\n${bullets.join('\n')}` : '';
+  }
+  return lead.replace(LEAD_IDENTITY_LINE, '').replace(LEAD_PROGRESS_PASSAGE, '').replace(LEAD_REPORT_PASSAGE, '');
+}
+
+// The agent-mode rule files (rules/agent-mode/): `conduct` (AGENT_MODE.md,
+// every agent-mode Lead), `mixy` (MIXY.md, Mixy only) and `routine`
+// (ROUTINE.md, schedule/webhook runs only).
+const AGENT_MODE_PARTS = {
+  conduct: ['AGENT_MODE.md', ''],
+  // Neutral heading: the user may rename Mixy (its identity line says the name).
+  mixy: ['MIXY.md', '# Your Role\n\n'],
+  routine: ['ROUTINE.md', '# Routine Run\n\n'],
+};
+
+function buildAgentModeContent({ PLUGIN_ROOT, part = 'conduct' }) {
+  const [file, heading] = AGENT_MODE_PARTS[part] || AGENT_MODE_PARTS.conduct;
+  const body = readOptional(path.join(PLUGIN_ROOT, 'rules', 'agent-mode', file)).trim();
+  return body ? `${heading}${body}` : '';
 }
 
-function buildLeadMetaContent({ PLUGIN_ROOT, DATA_DIR }) {
+function buildLeadMetaContent({ PLUGIN_ROOT, DATA_DIR, includeOutputStyle = true }) {
   const parts = [];
 
   const profilePreferences = buildProfilePreferencesContent(DATA_DIR);
@@ -382,7 +429,8 @@ function buildLeadMetaContent({ PLUGIN_ROOT, DATA_DIR }) {
   // (workflow, role, persona, memory, session lines) sits between it and the
   // first reply.
 
-  const outputStyle = loadOutputStyle({ PLUGIN_ROOT, DATA_DIR });
+  // Agent mode answers by AGENT_MODE.md alone; the coding Output Style stays out.
+  const outputStyle = includeOutputStyle ? loadOutputStyle({ PLUGIN_ROOT, DATA_DIR }) : '';
   if (outputStyle) parts.push(outputStyle);
 
   return parts.join('\n\n');
@@ -430,6 +478,7 @@ module.exports = {
   buildSharedToolContent,
   buildRouteRulesContent,
   buildLeadRoleContent,
+  buildAgentModeContent,
   buildLeadMetaContent,
   buildLeadLanguageContent,
   buildAgentRoleContent,

package/src/output-styles/common.md

@@ -7,15 +7,31 @@ partial: true
 
 ## Shared Output Format
 
-- Lead with the answer; no restated request or closing recap. Natural,
+- Lead with the answer; no restated request, closing recap, or offer. Natural,
   complete sentences in the reader's vocabulary, each point once; no arrow
   chains or unexplained shorthand.
-- Plain prose for simple questions; headings, lists, or tables only for real
-  structure. GitHub-flavored Markdown, `$$…$$` for display math.
 - Distinguish facts, inference, and uncertainty; keep exact literals and
   material risks. Caveats only when they change the reader's next step.
-- Never pad: no tangents, unsolicited offers, raw dumps, or process recaps.
-  Explicit depth requests win; depth applies to substantive answers, not the
-  one-line pre-tool briefing.
+- Show only what the reader needs: the decision, the result, the blocker, and
+  the action. No tangents, raw dumps, or process narration. Explicit depth
+  requests win; depth applies to substantive answers, not the one-line
+  pre-tool briefing.
 - At every depth keep decision-relevant failures, security warnings, and
   destructive-action confirmations; quote needed error text exactly.
+
+Layout (GitHub-flavored Markdown, `$$…$$` for display math): the content picks
+the format at every depth; depth sets only how much is said.
+
+- Mixed content gets one short block per kind (result, cause, finding,
+  decision needed, remaining work) under a plain label in the reader's
+  language, never one running paragraph; what the reader must decide or do
+  comes last, on its own.
+- A single answer or cause is short prose paragraphs without labels. Three or
+  more parallel items are a flat one-line list, numbered only for order, never
+  nested. A comparison across items is a small table with short cells.
+- Commands, paths, identifiers, and literals go in code; multi-line code,
+  diffs, and output in fenced blocks.
+- Bold only what the reader scans for: the decision or warning they must not
+  miss, a list item's leading label, the decisive number, the winning option;
+  never a sentence or a heading substitute. Headings only for long
+  multi-section answers; no emoji or horizontal rules.