thoth-agents 0.4.2 → 0.5.0

package/LICENSE

@@ -19,3 +19,33 @@ AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
 LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
 OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 SOFTWARE.
+
+---
+
+Third-party notices
+
+src/cli/pi-git-source.ts contains portions ported from
+@earendil-works/pi-coding-agent 1.0.2 (dist/utils/git.js), distributed under
+the following license:
+
+MIT License
+
+Copyright (c) 2025 Mario Zechner
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

package/README.md

@@ -1,5 +1,5 @@
 <div align="center">
-  <img src="img/thoth-agents-header.webp" alt="Seven cyber-Egyptian specialists led by Thoth, the Orchestrator" width="100%">
+  <img src="img/thoth-agents-header.webp" alt="Five cyber-Egyptian specialists led by Thoth, the Orchestrator" width="100%">
   <h1>Thoth-Agents</h1>
   <p><b>One conversation. The right specialists. A workflow that fits the task.</b></p>
   <p>Adaptive agent orchestration for OpenCode, Codex, Claude Code, and Pi.</p>
@@ -19,23 +19,28 @@
   </p>
 </div>
 
+> [!WARNING]
+> Thoth-Agents is under active development. Core concepts, workflows, and
+> specifications are still evolving, and significant breaking changes may occur
+> before a stable release.
+
 ---
 
 ## Why thoth-agents
 
 Describe what you want to build or fix. Thoth keeps the conversation together,
-handles straightforward work directly, and brings in specialists when research,
-design, implementation, or an independent review would help.
+directs specialists to discover and implement by default, and retains your goals,
+constraints, decisions, acceptance, and final synthesis in one root thread.
 
 Small changes stay small. Larger changes get a specification, a plan, and
 verification you can follow—without manually coordinating every agent.
 
-- **A team, not seven conversations.** One adaptive Orchestrator coordinates six
+- **A team, not six conversations.** One adaptive Orchestrator coordinates five
   specialists and brings their results back to you.
-- **The right amount of process.** Choose a quick Direct path or a structured
-  specification-driven development (SDD) workflow for more involved work.
-- **Specialists where they add value.** Repository discovery, current documentation,
-  UI/UX, focused edits, complex implementation, and independent review have distinct roles.
+- **Proportional process.** Small, clear, low-risk work stays artifact-free;
+  substantial work uses one ID-named record and proportionate verification.
+- **Specialists execute by default.** Repository discovery, current documentation,
+  UI/UX, bounded implementation, and independent review have distinct roles.
 - **Models you can tune.** Configure models per role to suit your workflow and
   the providers available in your harness.
 - **Continuity between sessions.** Published installs include setup of
@@ -44,9 +49,10 @@ verification you can follow—without manually coordinating every agent.
   memory lifecycle, persistence, and storage; thoth-agents only invokes its setup.
 
 > [!NOTE]
-> OpenCode is the default and most integrated path. 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.
+> Pi is the default harness and our recommendation for the best Thoth-Agents
+> experience. 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.
 
 ## Install
 
@@ -56,10 +62,10 @@ The commands below install at **global/user scope**.
 
 | Harness | What you get | Install command |
 | --- | --- | --- |
-| <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** | Native plugin, agent team, workflow skills, and memory setup. **Recommended starting point.** | `npx thoth-agents@latest install --agent=opencode` |
+| <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** | Native plugin, agent team, workflow skills, 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. **Run the prerequisites below first.** | `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, six specialists, delegation and research extensions, workflow skills, and memory setup. | `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 and research extensions, workflow skills, and memory setup. **Recommended for the best experience.** | `npx thoth-agents@latest install --agent=pi` |
 
 ### Claude Code prerequisites
 
@@ -75,10 +81,25 @@ claude plugin install thoth-agents@thoth-plugins --scope user
 > without writing changes. After installation, restart your harness; Claude Code
 > also supports `/reload-plugins`.
 
+Inside Pi, use `/subagents-model` to edit model and effort profiles for global
+and project subagent definitions. See [subagent profile configuration](docs/installation.md#configure-subagent-model-profiles-inside-pi).
+Use `/subagents-tools` to edit each specialist's explicit tool list, including
+registered active and inactive extension and MCP tools. Globs (including `*`) are
+manual advanced selections over all registered root tools; the panel preserves
+them read-only. Edit `disallowed_tools` manually for injected tools and trimming
+globs; only Oracle denies `ask_orchestrator` by default. See
+[tool selection controls](docs/installation.md#configure-specialist-tools-inside-pi).
+
 Pi setup currently supports the default `~/.pi/agent` root. See the
 [Pi installation guide](docs/installation.md#pi) for runtime requirements,
-existing-package conflicts, and recovery. Pi extensions run with your user's
-system permissions; agent tool allowlists are not an OS sandbox.
+existing-package conflicts, and recovery. Thoth manages its separate Pi
+delegation runtime as package `@thoth-agents/pi-subagents` (`0.1.0`) through
+`npm:@thoth-agents/pi-subagents@>=0.1.0`. Existing `pi-subagents` and
+`pi-subagents-j0k3r` installs need manual recovery through Pi's package manager
+before setup. Local checkout development uses `pnpm run setup:pi:local`, which
+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.
 
 For scopes, troubleshooting, or local checkout installation, see the
 [installation guide](docs/installation.md). Local Pi checkout installs keep
@@ -109,31 +130,30 @@ Open your repository in the harness and invoke the installed `thoth-init` skill:
 | Claude Code | `/thoth-agents:thoth-init` |
 | Pi | Ask: `Use the thoth-init skill to initialize this repository.` |
 
-This prepares the repository's `openspec/` governance for structured workflows.
-It does not install plugins or dependencies, and it preserves existing
-constitutions.
+This creates only missing minimum `.thoth/` governance, including
+`.thoth/constitution.md`, `.thoth/specs/`, and the change archive. It does not
+install plugins or dependencies and preserves existing project-owned governance
+and historical records.
 
 ### 3. Give Thoth a task
 
 Start with a goal, not a list of agents to manage. For example:
 
 ```text
-Fix the broken documentation link using the Direct route.
+Fix the broken documentation link with the smallest sufficient workflow.
 ```
 
 ```text
-Add CSV export to the reports page using Accelerated SDD.
-Keep the existing filters and include tests for empty results.
+Add CSV export to the reports page. Understand the existing filters, specify how empty results behave, and clarify any material decision before a test-first implementation. Keep the work proportional.
 ```
 
 ```text
-Use Full SDD to plan a migration from our current authentication system.
-Explore the risks before proposing changes.
+Plan a migration from our current authentication system. Explore the risks, specify acceptance, and clarify material decisions before implementation. If classification is substantial, keep the plan and acceptance in one `.thoth/changes/<id>/<id>.md` record.
 ```
 
-You can name a route explicitly or let Thoth recommend one. Describe your
-constraints and expected outcome; the Orchestrator decides whether to handle
-work directly or bring in a specialist.
+Describe the desired outcome, constraints, and acceptance. The Orchestrator
+first builds proportional understanding, then classifies by coordination,
+uncertainty, and risk and selects the fitting specialist.
 
 ## Meet the team
 
@@ -142,7 +162,7 @@ work directly or bring in a specialist.
 <table>
   <tr>
     <td width="25%" align="center"><img src="img/agents/orchestrator.webp" width="160" alt="Thoth as the Orchestrator"></td>
-    <td><b>Orchestrator · Keeps the work moving</b><br><br>Your main point of contact. Understands the goal, recommends a workflow, handles bounded work, and coordinates specialists without handing you the management overhead.</td>
+    <td><b>Orchestrator · Keeps the work moving</b><br><br>Your main point of contact. Retains goals, constraints, decisions and acceptance while directing specialists, with a bounded exception for known-source consultation or minimal low-risk edits.</td>
   </tr>
 </table>
 
@@ -155,41 +175,40 @@ work directly or bring in a specialist.
 
 ### Design and implementation
 
-| Designer | Quick | Deep |
-| :---: | :---: | :---: |
-| <img src="img/agents/designer.webp" width="150" alt="Hathor as the Designer"> | <img src="img/agents/quick.webp" width="150" alt="Horus as Quick"> | <img src="img/agents/deep.webp" width="150" alt="Sobek as Deep"> |
-| **Makes interfaces work well.** Owns UI/UX, accessibility, implementation, and visual quality. | **Makes focused changes.** Handles clear, narrow, low-risk implementation tasks. | **Handles complex changes.** Works through coupled behavior, edge cases, and correctness-critical implementation. |
+| Designer | Worker |
+| :---: | :---: |
+| **Makes interfaces work well.** Owns material UI/UX, accessibility, interaction, and visual quality. | **Implements changes.** Owns delegated implementation, including coupled behavior, edge cases, migrations, and correctness-critical work. |
 
 Research and review specialists are read-only. Implementation work has one
 writer per area; independent areas can proceed in parallel when the harness
 supports it. You do not need to summon every role for every task.
 
-## Choose your workflow
-
-SDD means **specification-driven development**: agree on the intended result,
-plan the work, implement it, and check it against that intent.
-
-| Route | Best for | What to expect |
-| --- | --- | --- |
-| **Direct** | Clear, bounded, low-risk fixes and documentation changes. | Implement → verify. No planning artifacts required. |
-| **Accelerated** | Features spanning several areas, moderate risk, or a request to use SDD. | Specify → plan → tasks → implement → verify → archive. Planning runs in one pass unless a material decision needs you. |
-| **Full** | Uncertain requirements, architectural changes, or high-cost failures. | Explore first, then follow the structured workflow with separate planning checkpoints. |
-
-For Accelerated and Full, the specification, plan, tasks, and verification
-reports live under `openspec/`, so you can inspect what was agreed and what
-was checked. You can choose an optional Oracle plan review before implementation;
-final verification is required either way. These routes and materially risky
-Direct work use an independent Oracle for final verification.
-
-> [!TIP]
-> You stay in control of the route. Say “Use Direct,” “Use Accelerated SDD,” or
-> “Use Full SDD” when you already know how much structure you want.
-
-The installed skills cover project initialization, SDD, project principles,
-plan review, and archiving. External skills add test-driven development,
-behavior-preserving simplification, focused repository context, and architectural
-questioning when needed. See [Skills and MCPs](docs/skills-and-mcps.md) and the
-[SDD workflow guide](docs/sdd-pipeline.md) for details.
+## Proportional SDD
+
+Every change completes explore, specify, and clarify in order, proportionally to
+its uncertainty and impact. These steps do not force a document, specialist, or
+interview. The Orchestrator resolves repository facts through evidence and asks
+only when a material human-owned decision cannot safely be inferred; unresolved
+material uncertainty blocks classification.
+
+Only after understanding does it classify by meaningful coordination and
+contract impact, uncertainty, and risk. File count alone does not increase scope:
+a clear, low-risk localized mechanical change may touch several files and remain
+small. Small work uses test-first implementation and focused verification with
+no record. Substantial or materially risky work plans in one
+`.thoth/changes/<id>/<id>.md` record after classification; risk may warrant
+planning even for a patch-sized change. No alias, duplicate record, report,
+evidence directory, or process tooling is created.
+
+For substantial work, optional fresh Oracle plan review runs only when selected.
+After a selected `[OKAY]`, root separately asks Implement (Recommended) or Stop;
+review alone never authorizes implementation or replaces final verification.
+Substantial and materially risky work receives a fresh read-only Oracle
+verification. Declared durable changes sync transactionally to
+`.thoth/specs/` after PASS. Active governance lives at `.thoth/constitution.md`;
+historical records remain preserved, and provider memory stays separate. Native
+harness execution and liveness remain authoritative. See the [SDD guide](docs/sdd-pipeline.md)
+and [Skills and MCPs](docs/skills-and-mcps.md) for limits.
 
 ## Configure and update
 
@@ -225,6 +244,15 @@ skills and provider setup—not just the plugin. Native marketplace updates alon
 do not prove those other pieces are current. Use `status` to inspect the last
 complete CLI-managed installation and follow any reported recovery actions.
 
+For Pi, the first-party `thoth-agents` package remains exact and receipt-verified.
+The five mandatory external extensions use stable minimum-only `>=` ranges, so
+Pi's native package manager can update them independently without waiting for a
+Thoth release. Status validates each installed manifest's package name and
+SemVer floor; newer stable versions are healthy, while prerelease, malformed,
+missing, or older versions are not. Re-running Install or applying Update
+migrates legacy exact external sources through Pi while preserving package
+resource filters and unrelated settings.
+
 ## Documentation
 
 ### User guides
@@ -233,7 +261,7 @@ complete CLI-managed installation and follow any reported recovery actions.
 | --- | --- |
 | [Installation](docs/installation.md) | Check prerequisites, preview setup, troubleshoot, and repair an installation. |
 | [Quick Reference](docs/quick-reference.md) | Find commands, roles, skills, and workflow reminders. |
-| [SDD Pipeline](docs/sdd-pipeline.md) | Understand planning, review, verification, and archiving. |
+| [SDD pipeline](docs/sdd-pipeline.md) | Understand routes, planning, review, verification, and archiving. |
 | [Skills and MCPs](docs/skills-and-mcps.md) | See the included workflows, research tools, and memory boundaries. |
 | [Provider Configuration](docs/provider-configurations.md) | Configure models and providers. |
 | [Codex Install](docs/codex-install.md) | Follow Codex-specific setup, activation, and trust requirements. |

package/dist/agents/prompt-sections.d.ts

@@ -4,11 +4,12 @@ import type { ModelEntry } from './prompt-utils';
 type ModelFamily = 'openai';
 export type SemanticMemoryAccess = 'dispatch-scoped';
 export type ReadOnlyAgentRole = 'explorer' | 'librarian' | 'oracle';
-export type WriteCapableAgentRole = 'designer' | 'quick' | 'deep';
+export type WriteCapableAgentRole = 'designer' | 'worker';
 export interface QuestionProtocolSection {
     kind: 'question-protocol';
     toolConcept: 'userQuestion';
     audience: 'root' | 'child';
+    role?: AgentRoleName;
 }
 export interface SubagentRulesSection {
     kind: 'subagent-rules';
@@ -25,6 +26,7 @@ export interface ResponseBudgetSection {
 export interface StepBudgetSection {
     kind: 'step-budget';
     steps: number;
+    role?: string;
 }
 export interface ModelFamilySection {
     kind: 'model-family';
@@ -40,11 +42,11 @@ export type RolePromptSection = PromptSection;
 export interface PromptSectionRenderer<TSection extends PromptSection> {
     render(section: TSection, dialect: HarnessPromptDialect): string;
 }
-export declare function createQuestionProtocolSection(audience?: 'root' | 'child'): QuestionProtocolSection;
+export declare function createQuestionProtocolSection(audience?: 'root' | 'child', role?: AgentRoleName): QuestionProtocolSection;
 export declare function createSubagentRulesSection(memoryAccess?: SemanticMemoryAccess): SubagentRulesSection;
 export declare function createReasoningDisciplineSection(): ReasoningDisciplineSection;
 export declare function createResponseBudgetSection(): ResponseBudgetSection;
-export declare function createStepBudgetSection(steps?: number): StepBudgetSection | undefined;
+export declare function createStepBudgetSection(steps?: number, role?: string): StepBudgetSection | undefined;
 export declare function detectModelFamilyFromModel(model?: string | ModelEntry[]): ModelFamily | undefined;
 export declare function createModelFamilySection(role: AgentPromptRole, model?: string | ModelEntry[]): ModelFamilySection | undefined;
 export declare function createOrchestratorPromptSections(): RolePromptSection[];

package/dist/agents/prompt-utils.d.ts

@@ -15,7 +15,7 @@ export declare const QUESTION_PROTOCOL: string;
 export declare const SUBAGENT_RULES: string;
 export declare const RESPONSE_BUDGET: string;
 export declare function appendPromptSections(...sections: Array<string | undefined>): string;
-export declare function getStepBudgetPromptSection(steps?: number): string | undefined;
+export declare function getStepBudgetPromptSection(steps?: number, role?: string): string | undefined;
 export declare function getModelFamilyPromptSection(role: AgentPromptRole, model?: string | ModelEntry[]): string | undefined;
 export declare function replacePromptPlaceholders(template: string, placeholders?: Record<string, string | number | undefined>): string;
 export declare function composeAgentPrompt({ basePrompt, customPrompt, customAppendPrompt, placeholders, }: ComposeAgentPromptOptions): string;

package/dist/agents/worker.d.ts

@@ -0,0 +1,2 @@
+import type { AgentDefinition } from './orchestrator';
+export declare function createWorkerAgent(model: string, customPrompt?: string, customAppendPrompt?: string): AgentDefinition;

package/dist/{chunk-STCURXAQ.js → chunk-MRFCA3U2.js}

@@ -16,34 +16,7 @@ import {
   loadAgentPrompt,
   renderAgentRoutingDescription,
   renderRolePrompt
-} from "./chunk-EFGSGPYM.js";
-
-// src/agents/deep.ts
-var DEEP_PROMPT = renderRolePrompt(
-  createWriteCapableSpecialistPromptSections("deep"),
-  OPENCODE_PROMPT_DIALECT
-);
-function createDeepAgent(model, customPrompt, customAppendPrompt) {
-  const prompt = composeAgentPrompt({
-    basePrompt: DEEP_PROMPT,
-    customPrompt,
-    customAppendPrompt: appendPromptSections(
-      getModelFamilyPromptSection("deep", model),
-      customAppendPrompt
-    )
-  });
-  return {
-    name: "deep",
-    description: "Synchronous write-capable implementation agent optimized for thorough context analysis, edge cases, and correctness \u2014 not for bulk mechanical changes.",
-    config: {
-      model,
-      temperature: 0.1,
-      prompt,
-      color: "secondary"
-      // steps: 80,
-    }
-  };
-}
+} from "./chunk-OW6I7ZUU.js";
 
 // src/agents/designer.ts
 var DESIGNER_PROMPT = renderRolePrompt(
@@ -61,7 +34,7 @@ function createDesignerAgent(model, customPrompt, customAppendPrompt) {
   });
   return {
     name: "designer",
-    description: "Synchronous write-capable UI/UX implementation agent with ownership of approach, execution, and visual verification.",
+    description: "Write-capable UI/UX implementation agent with ownership of approach, execution, and visual verification.",
     config: {
       model,
       temperature: 0.4,
@@ -140,7 +113,7 @@ function createOracleAgent(model, customPrompt, customAppendPrompt) {
   });
   return {
     name: "oracle",
-    description: "Synchronous read-only strategic advisor for debugging, architecture, code review, and SDD plan review.",
+    description: "Read-only strategic advisor for debugging, architecture, focused plan review, and independent verification.",
     config: {
       model,
       temperature: 0.1,
@@ -166,7 +139,7 @@ function createOrchestratorAgent(model, customPrompt, customAppendPrompt) {
   });
   const definition = {
     name: "orchestrator",
-    description: "Adaptive root coordinator that handles bounded work directly and delegates only for net gain.",
+    description: "Director root that retains decisions and acceptance while specialists discover and implement by default.",
     config: {
       temperature: 0.1,
       prompt,
@@ -184,29 +157,29 @@ function createOrchestratorAgent(model, customPrompt, customAppendPrompt) {
   return definition;
 }
 
-// src/agents/quick.ts
-var QUICK_PROMPT = renderRolePrompt(
-  createWriteCapableSpecialistPromptSections("quick"),
+// src/agents/worker.ts
+var WORKER_PROMPT = renderRolePrompt(
+  createWriteCapableSpecialistPromptSections("worker"),
   OPENCODE_PROMPT_DIALECT
 );
-function createQuickAgent(model, customPrompt, customAppendPrompt) {
+function createWorkerAgent(model, customPrompt, customAppendPrompt) {
   const prompt = composeAgentPrompt({
-    basePrompt: QUICK_PROMPT,
+    basePrompt: WORKER_PROMPT,
     customPrompt,
     customAppendPrompt: appendPromptSections(
-      getModelFamilyPromptSection("quick", model),
+      getModelFamilyPromptSection("worker", model),
       customAppendPrompt
     )
   });
   return {
-    name: "quick",
-    description: "Synchronous write-capable implementation agent optimized for fast, mechanical, well-bounded changes \u2014 including uniform patterns across multiple files.",
+    name: "worker",
+    description: "Write-capable agent for bounded nonvisual implementation regardless of complexity, with thorough context analysis, edge-case handling, and correctness verification.",
     config: {
       model,
-      temperature: 0.2,
+      temperature: 0.1,
       prompt,
-      color: "success"
-      // steps: 30,
+      color: "secondary"
+      // steps: 80,
     }
   };
 }
@@ -295,24 +268,7 @@ var BUILTIN_PERMISSION_PRESETS = {
       "~/.config/opencode/skills/**": "allow"
     }
   },
-  quick: {
-    read: "allow",
-    edit: "allow",
-    glob: "allow",
-    grep: "allow",
-    list: "allow",
-    bash: "allow",
-    question: "allow",
-    codesearch: "allow",
-    lsp: "allow",
-    skill: "allow",
-    todowrite: "deny",
-    task: "deny",
-    external_directory: {
-      "~/.config/opencode/skills/**": "allow"
-    }
-  },
-  deep: {
+  worker: {
     read: "allow",
     edit: "allow",
     glob: "allow",
@@ -357,7 +313,10 @@ function applyOverrides(agent, override) {
   }
 }
 function applyStepBudgetPrompt(agent) {
-  const stepBudgetPrompt = getStepBudgetPromptSection(agent.config.steps);
+  const stepBudgetPrompt = getStepBudgetPromptSection(
+    agent.config.steps,
+    agent.name
+  );
   if (!stepBudgetPrompt) {
     return;
   }
@@ -398,8 +357,7 @@ var SUBAGENT_FACTORIES = {
   librarian: createLibrarianAgent,
   oracle: createOracleAgent,
   designer: createDesignerAgent,
-  quick: createQuickAgent,
-  deep: createDeepAgent
+  worker: createWorkerAgent
 };
 function createAgents(config) {
   const protoSubAgents = Object.entries(SUBAGENT_FACTORIES).map(([name, factory]) => {
@@ -532,7 +490,8 @@ function spawn(command, options = {}) {
       options.stdout ?? "pipe",
       options.stderr ?? "pipe"
     ],
-    shell: options.shell
+    shell: options.shell,
+    windowsHide: options.windowsHide ?? true
   });
   const managed = {
     stdin: child.stdin ?? fallbackStdin(),
@@ -564,7 +523,8 @@ function spawnSync(command, options = {}) {
       options.stdout ?? "pipe",
       options.stderr ?? "pipe"
     ],
-    shell: options.shell
+    shell: options.shell,
+    windowsHide: options.windowsHide ?? true
   });
   return { exitCode: result.status };
 }
@@ -659,7 +619,10 @@ async function isOpenCodeInstalled() {
   for (const opencodePath of paths) {
     try {
       const invocation = getOpenCodeVersionInvocation(opencodePath);
-      const proc = spawn(invocation.command, invocation.options);
+      const proc = spawn(invocation.command, {
+        ...invocation.options,
+        windowsHide: true
+      });
       let timer;
       const timedOut = /* @__PURE__ */ Symbol("timeout");
       try {
@@ -694,7 +657,10 @@ async function getOpenCodeVersion() {
   const opencodePath = resolveOpenCodePath();
   try {
     const invocation = getOpenCodeVersionInvocation(opencodePath);
-    const proc = spawn(invocation.command, invocation.options);
+    const proc = spawn(invocation.command, {
+      ...invocation.options,
+      windowsHide: true
+    });
     const output = await new Response(proc.stdout).text();
     await proc.exited;
     if (proc.exitCode === 0) {