thoth-agents 0.13.0 → 0.13.2

package/README.md

@@ -52,8 +52,8 @@ verification you can follow—without manually coordinating every agent.
   successful; see [provider outcomes](docs/installation.md#thoth-mem-companion-setup).
 
 > [!NOTE]
-> OpenCode is the CLI's default harness. Pi is our recommendation for the best
-> Thoth-Agents experience. All four harnesses share the workflow and role design,
+> OpenCode is the CLI's default harness, fully supported on OpenCode v2 (`>=2.2.6`).
+> All four harnesses share the workflow and role design,
 > but their permissions, delegation, and runtime capabilities are not identical.
 > Your harness's trust and approval rules still apply.
 
@@ -68,7 +68,7 @@ The commands below install at **global/user scope**.
 | <a href="https://github.com/anomalyco/opencode"><picture><source media="(prefers-color-scheme: dark)" srcset="https://svgl.app/library/opencode-dark.svg"><img src="https://svgl.app/library/opencode.svg" alt="OpenCode logo" width="48" height="48"></picture></a><br>**OpenCode** | OpenCode >=2.2.6 native v2 plugin, package skills, agent team, and memory setup. | `npx thoth-agents@latest install --agent=opencode` |
 | <a href="https://github.com/openai/codex"><img src="https://github.com/openai.png?size=120" alt="OpenAI logo — Codex" width="48" height="48"></a><br>**Codex** | Native plugin plus the required global agent and instruction setup. **Close Codex first.** | `npx thoth-agents@latest install --agent=codex` |
 | <a href="https://claude.com/product/claude-code"><img src="https://github.com/anthropics.png?size=120" alt="Anthropic logo — Claude Code" width="48" height="48"></a><br>**Claude Code** | Marketplace agents and skills, completed by the CLI's external skills and memory setup. | `npx thoth-agents@latest install --agent=claude` |
-| <a href="https://github.com/earendil-works/pi"><img src="https://raw.githubusercontent.com/EremesNG/thoth-mem/master/img/pi.svg" alt="Pi logo" width="48" height="48"></a><br>**Pi** | Native package, five specialists, delegation, research, theme, sidebar and background-task extensions, workflow skills, and memory setup. **Recommended for the best experience.** | `npx thoth-agents@latest install --agent=pi` |
+| <a href="https://github.com/earendil-works/pi"><img src="https://raw.githubusercontent.com/EremesNG/thoth-mem/master/img/pi.svg" alt="Pi logo" width="48" height="48"></a><br>**Pi** | Native package, five specialists, delegation, research, theme, sidebar and background-task extensions, workflow skills, and memory setup. | `npx thoth-agents@latest install --agent=pi` |
 
 Codex hooks require your trust; installation does not silently grant it. Consent
 during setup, or review and trust them in Codex via `/hooks` or the startup
@@ -109,7 +109,8 @@ direct calls, independently controlled by `enable_ask_orchestrator` and
 `enable_notify_orchestrator` (default true, global/project cascade) and denials. See
 [tool selection controls](docs/installation.md#configure-specialist-tools-inside-pi).
 
-Pi setup currently supports the default `~/.pi/agent` root. See the
+Pi setup requires **Pi `>=1.5.0`** and currently supports the default
+`~/.pi/agent` root. See the
 [Pi installation guide](docs/installation.md#pi) for runtime requirements,
 existing-package conflicts, and recovery. Thoth manages its separate Pi
 delegation runtime as package `@thoth-agents/pi-subagents` through
@@ -122,6 +123,19 @@ points Pi at the fork under `pi-packages/pi-subagents`; publishing the fork is
 not required for that path. Pi extensions run with your user's system
 permissions; agent tool allowlists are not an OS sandbox.
 
+### Optional OpenCode multi-account plugin (manual install)
+
+[`@thoth-agents/opencode-multi-codex`](opencode-packages/opencode-multi-codex/README.md)
+lets you use ChatGPT accounts concurrently: account 1 uses `openai`,
+with additional accounts exposed as numbered providers such as `openai-codex-2`.
+Pin agents to different accounts, inspect quota, and rotate around usage-limit
+429s. Manage accounts with `/codex-add`, `/codex-remove`, `/codex-switch`, and
+`/codex-status`.
+
+Install it separately using the package guide; the thoth-agents CLI does **not**
+install or manage it. It requires OpenCode's legacy ChatGPT/Codex OAuth login;
+API keys and the newer ChatGPT token-sharing flow are not supported.
+
 ### Optional Pi extensions (manual install)
 
 The installer does **not** install or manage the Claude/Antigravity bridges,
@@ -150,7 +164,7 @@ for both `claude-bridge` and `antigravity`. See
 
 ### Pi package inventory
 
-The eleven first-party `@thoth-agents/pi-*` packages are:
+The twelve first-party `@thoth-agents/pi-*` packages are:
 
 | Package | Setup |
 | --- | --- |
@@ -158,6 +172,7 @@ The eleven first-party `@thoth-agents/pi-*` packages are:
 | [`@thoth-agents/pi-askcodex`](pi-packages/pi-askcodex/README.md) | Optional/manual |
 | [`@thoth-agents/pi-background-tasks`](pi-packages/pi-background-tasks/README.md) | Installer-managed |
 | [`@thoth-agents/pi-claude-bridge`](pi-packages/pi-claude-bridge/README.md) | Optional/manual |
+| [`@thoth-agents/pi-contained-shell`](pi-packages/pi-contained-shell/README.md) | Installer-managed |
 | [`@thoth-agents/pi-core`](pi-packages/pi-core/README.md) | Shared library; not installer-managed |
 | [`@thoth-agents/pi-openai-fast`](pi-packages/pi-openai-fast/README.md) | Optional/manual |
 | [`@thoth-agents/pi-questions-user`](pi-packages/pi-questions-user/README.md) | Installer-managed |
@@ -290,6 +305,14 @@ Use the interactive CLI to inspect setup and configure role models:
 npx thoth-agents@latest
 ```
 
+Other CLI commands (run with `npx thoth-agents@latest <command>`):
+
+- `list`: list managed surfaces and actions.
+- `sync`: preview managed configuration sync.
+- `model`: preview role model/provider settings.
+
+Add `--apply` to apply a reviewed `sync` or `model` plan.
+
 Choose models your harness and provider account can access. OpenCode ships the
 **`openai` model profile**. Run `/subagents-model` inside OpenCode to edit named
 model/variant profiles for all six roles, including orchestrator. Selection is
@@ -329,11 +352,13 @@ diagnostics. To remove attributable Codex surfaces, preview with
 to apply.
 
 For Pi, the first-party `thoth-agents` package remains exact and receipt-verified.
-The nine selected extensions use stable minimum-only `>=` ranges, so
+The nine selected external packages use stable minimum-only `>=` ranges, so
 Pi's native package manager can update them independently without waiting for a
 Thoth release. Install and applied Update now include
 `@thoth-agents/pi-thoth-theme`, `@thoth-agents/pi-background-tasks` and
-`@thoth-agents/pi-sidebar` at `>=0.3.0`.
+`@thoth-agents/pi-sidebar` at `>=0.3.0`, plus
+`@thoth-agents/pi-contained-shell` at `>=0.1.0` for Windows shell containment
+(a no-op on other platforms).
 Existing copies of these three packages from any source are preserved and verified
 without reinstalling when compatible; older copies are left untouched and block
 completion with manual upgrade guidance. Ambiguous sources fail closed. Status
@@ -350,17 +375,27 @@ The status line uses `thoth:subagents:usage`, not the removed raw
 `>=0.3.0` together and `/reload` to retain subagent cost display; the usage
 checkpoint discriminator is unchanged.
 
+In Pi, `/thoth-theme` opens settings for icons, animation intensity, tool density,
+and module toggles. Visual settings apply live; module toggles require `/reload`.
+Opt-in `tools.collapsed` settings (`frame` | `compact` | `line`) let you keep framed
+cards or use compact tool rows, including subagent and todo cards. Expanded output
+is unchanged. See the [theme guide](pi-packages/pi-thoth-theme/README.md).
+
 The shared Pi Work panel uses pi-core's v2 data-only provider contract and
 [discovery/action API](pi-packages/pi-core/README.md#discovery-and-actions).
 Upgrade pi-subagents, pi-background-tasks and pi-todo together to `>=0.3.0`, then
 `/reload`; first-owner arbitration hides incompatible sections in mixed v1/v2
 installs. Subagent state is v2; background state, usage, requests and Render KIT remain v1.
+Todos support grouping, ordering, and archiving in compact views; completed-row
+limits preserve pending work. A compact Work summary remains visible when another
+component replaces the editor.
 
 The [Pi sidebar](pi-packages/pi-sidebar/README.md) adds read-only Session,
 Workspace, Cost and discovered Todos/Subagents/Background panels in fullscreen and
 regular mode. `/sidebar` toggles, `/sidebar panels` lists ids for show/hide/order
-commands, and `/sidebar resize` resizes. `/sidebar settings` edits panel order,
-visibility, startup and default width; `/sidebar cost` opens task-cost curves
+commands. Todo rows share compact counts with Work, group by status, and respect
+archive and completed-row limits. `/sidebar resize` resizes. `/sidebar settings`
+edits panel order, visibility, startup and default width; `/sidebar cost` opens task-cost curves
 against elapsed time per task, labeled by display name or a short task-summary name.
 Themed Agents/Background headers show static status-icon counts, omitting zero
 counts (Todos keeps completed/total), and detail commands; Workspace shows a
@@ -374,8 +409,12 @@ state v2 adds up to 100 cost-ranked persisted session summaries in `history`. Pr
 have guarded fallback; regular mode covers the live viewport, not historical
 scrollback. Update related extensions together to avoid mixed old UI owners.
 
-Pi's question UI replaces the editor without covering chat. `Ctrl+]` collapses it
-to a one-line dock and returns input to the editor; during an active run, `Enter`
+Pi's question UI replaces the editor without covering chat. Panels size to content,
+capped at 45% of terminal height. Narrow previews start collapsed; `p` toggles them
+outside editors, and Alt+arrow keys scroll context. Tool `context` is limited to
+500 characters; longer proposals belong in the response before the question.
+`Ctrl+]` collapses the question UI to a one-line dock and returns input to the
+editor; during an active run, `Enter`
 queues a steering message without answering the question. `Ctrl+]` re-expands
 with answers and drafts preserved, and the open tool-call card shows collapsed
 state. While open, `Ctrl+]` shadows editor `jumpForward`; collapsed `Esc` belongs

package/dist/{chunk-YCWGTDFE.js → chunk-5ANFI4LW.js}

@@ -950,6 +950,17 @@ function detectCurrentConfig() {
 
 // src/config/loader.ts
 var PROMPTS_DIR_NAME = "thoth-agents";
+function withoutInjectedDefaults(parsed, raw) {
+  if (parsed === null || typeof parsed !== "object" || Array.isArray(parsed) || raw === null || typeof raw !== "object") {
+    return parsed;
+  }
+  return Object.fromEntries(
+    Object.entries(parsed).filter(([key]) => Object.hasOwn(raw, key)).map(([key, value]) => [
+      key,
+      withoutInjectedDefaults(value, raw[key])
+    ])
+  );
+}
 function loadConfigFromPath(configPath, unsupportedKeyPaths) {
   try {
     const content = fs2.readFileSync(configPath, "utf-8");
@@ -963,7 +974,7 @@ function loadConfigFromPath(configPath, unsupportedKeyPaths) {
       console.warn(result.error.format());
       return null;
     }
-    return result.data;
+    return withoutInjectedDefaults(result.data, rawConfig);
   } catch (error) {
     if (error instanceof Error && "code" in error && error.code !== "ENOENT") {
       console.warn(
@@ -988,9 +999,9 @@ function findConfigPath(basePath) {
 function deepMerge(base, override) {
   if (!base) return override;
   if (!override) return base;
-  const result = { ...base };
+  const result = Object.assign(/* @__PURE__ */ Object.create(null), base);
   for (const key of Object.keys(override)) {
-    const baseVal = base[key];
+    const baseVal = Object.hasOwn(base, key) ? base[key] : void 0;
     const overrideVal = override[key];
     if (typeof baseVal === "object" && baseVal !== null && typeof overrideVal === "object" && overrideVal !== null && !Array.isArray(baseVal) && !Array.isArray(overrideVal)) {
       result[key] = deepMerge(
@@ -1030,7 +1041,16 @@ function loadPluginConfig(directory) {
       `[thoth-agents] Config keys preset/presets are unsupported and ignored (${[...unsupportedKeyPaths].join(", ")}).`
     );
   }
-  return config;
+  const result = PluginConfigSchema.safeParse(config);
+  if (!result.success) {
+    const sourcePaths = [userConfigPath, projectConfigPath].filter(Boolean);
+    console.warn(
+      `[thoth-agents] Invalid merged config from ${sourcePaths.join(", ")}:`
+    );
+    console.warn(result.error.format());
+    return {};
+  }
+  return result.data;
 }
 function loadAgentPrompt(agentName) {
   const promptsDir = path2.join(getConfigDir(), PROMPTS_DIR_NAME);
@@ -1259,37 +1279,37 @@ var AGENT_ROLES = [
     mode: "write-capable",
     dispatch: "task",
     canMutateWorkspace: true,
-    scope: "UI/UX decisions, implementation, and visual verification",
-    responsibility: "Own user-facing implementation choices and visual quality for UI work.",
+    scope: "UI/UX and creative/artistic decisions, implementation, and verification",
+    responsibility: "Own UI/UX, creative/artistic output, and aesthetic quality.",
     useWhen: [
-      "User-facing UI/UX, interaction, accessibility, or visual quality is material."
+      "Material UI/UX, interaction, accessibility, visual quality, or creative/artistic work: visual style, motion/animation, layouts/levels/maps, characters, visual/audio assets, branding, creative copy, narrative."
     ],
     doNotUseWhen: [
-      "Not for backend-only, non-visual, or correctness-heavy cross-cutting work."
+      "Not for engineering-only or correctness-heavy cross-cutting work."
     ],
     escalateWhen: [
       "Escalate coupled contracts, migrations, or high risk to worker."
     ],
     toolGovernance: [
-      "may edit focused UI/UX files",
-      "owns screenshots and visual QA",
+      "may edit focused UI/UX and creative/artistic files",
+      "owns visual and creative QA, including screenshots",
       "does not delegate further"
     ],
-    verification: ["includes visual verification when applicable"]
+    verification: ["includes visual and creative verification when applicable"]
   },
   {
     name: "worker",
     mode: "write-capable",
     dispatch: "task",
     canMutateWorkspace: true,
-    scope: "bounded nonvisual implementation and verification",
-    responsibility: "Handle bounded nonvisual implementation with full local context, including exact low-risk edits and correctness-critical, multi-file, edge-case-heavy, or high-risk work.",
+    scope: "bounded engineering implementation and verification",
+    responsibility: "Handle bounded engineering implementation (business logic, backend, systems, mechanic scripts, calculations, AI) with full local context, including exact low-risk edits and correctness-critical, multi-file, edge-case-heavy, or high-risk work.",
     useWhen: [
-      "Known bounded nonvisual implementation is ready, regardless of complexity; root direct work is limited to the bounded implementation-ownership exceptions.",
+      "Known bounded engineering implementation, regardless of complexity; root direct work only under bounded implementation-ownership exceptions.",
       "Correctness-critical work may be multi-file, edge-case-heavy, migration, concurrency, shared-contract, or high-risk."
     ],
     doNotUseWhen: [
-      "Not for visual-only work, reviewed commits, or work explicitly retained by the user in root."
+      "Not for visual-only or artistic/creative work, reviewed commits, or work explicitly retained by the user in root."
     ],
     escalateWhen: ["Return product or architecture choices to root."],
     toolGovernance: [
@@ -1321,7 +1341,8 @@ var IMPLEMENTATION_OWNERSHIP_POLICY = {
     "Direct-work exceptions: known low-risk mechanical work (reviewed commits included), no rediscovery; minimal authorized low-risk edits with known scope/verification, no discovery/independent judgment."
   ],
   writerRouting: [
-    "Known bounded implementation: directly to designer/worker, no mandatory Explorer; librarian: external evidence; Oracle: independent judgment; no fixed all-role pipeline."
+    "Known bounded implementation: directly to designer/worker, no mandatory Explorer; librarian: external evidence; Oracle: independent judgment; no fixed all-role pipeline.",
+    "Split mixed creative/engineering units between designer and worker with an agreed interface. User/root spec owns system/game rules and balance; worker implements."
   ],
   evidenceHandling: [
     "Conclusions/localized evidence/uncertainty, not files/source/log dumps; next action only from Oracle/Worker/Designer. Never repeat delegated discovery; fill gaps via targeted requests or bounded identified-evidence inspection. Independent verification stays mandatory."
@@ -1442,7 +1463,7 @@ function cloneReturnContract(contract) {
 var VERIFICATION_PROTOCOL = [
   "Completion reports identify changed files and verification evidence.",
   "Behavior changes receive the smallest sufficient automated check or a declared manual check.",
-  "Visual changes receive designer-owned visual QA when applicable."
+  "Visual and creative output receives designer-owned visual and creative QA when applicable."
 ];
 var AGENT_PACK_CONTRACT = {
   roles: [...AGENT_ROLES],
@@ -2398,8 +2419,9 @@ var ROLE_SPECIFIC_RULES = {
     "Reject self-review: implementing root/writer cannot substitute for independent oracle judgment."
   ],
   designer: [
-    "Own user-facing choices, implementation, and visual verification.",
-    "Check relevant responsive and interaction states when feasible."
+    "Own UI/UX and creative/artistic choices, implementation, and verification.",
+    "Check relevant responsive and interaction states when feasible. Self-check creative assets for intent/style consistency.",
+    "Use the creative tools the harness or project provides; report a capability gap otherwise."
   ],
   worker: [
     "Start at supplied entrypoints; read further only to resolve a concrete missing fact. Use tests first for behavior changes.",
@@ -2557,7 +2579,7 @@ function getRoleModelProfile(role) {
     case "oracle":
       return "Challenge assumptions and return evidence-backed judgment.";
     case "designer":
-      return "Make concrete UX choices and verify the visible result.";
+      return "Make concrete UX and creative choices; verify intent/style consistency.";
     case "worker":
       return "Trace shared behavior, test assumptions, and verify edge cases.";
   }

package/dist/{chunk-VVYAQTYE.js → chunk-7B7JYFIJ.js}

@@ -19,15 +19,14 @@ import {
   renderAgentRoutingDescription,
   renderPromptSection,
   renderRolePrompt
-} from "./chunk-YCWGTDFE.js";
+} from "./chunk-5ANFI4LW.js";
 
 // src/harness/writers/pi-agent.ts
 var PI_MANAGED_OWNER = "thoth-agents";
 var PI_ROOT_START = "<!-- thoth-agents:pi-root:start -->";
 var PI_ROOT_END = "<!-- thoth-agents:pi-root:end -->";
 var LIBRARIAN_RESEARCH_TOOLS = [
-  "resolve-library-id",
-  "query-docs",
+  "mcp__context7",
   "mcp",
   "web_search",
   "fetch_content",
@@ -855,7 +854,7 @@ function diagnostics() {
       code: "pi.mcp.adapter-backed",
       harness: "pi",
       surface: "grep.app",
-      message: "Pi has no generic native MCP client in this contract; only grep.app is exposed through pi-mcp-adapter, while Context7 and pi-web-access remain native extensions.",
+      message: "Pi has no generic native MCP client in this contract; grep.app and Context7 are exposed through pi-mcp-adapter, while pi-web-access remains a native extension.",
       fallback: "diagnostic-only"
     }
   ];