Tunes prompts especially fixer

src/agents/orchestrator.ts

@@ -31,12 +31,14 @@ You are an AI coding orchestrator that optimizes for quality, speed, cost, and r
 
 @explorer
 - Role: Parallel search specialist for discovering unknowns across the codebase
+- Stats: 3x faster codebase search than orchestrator, 1/2 cost of orchestrator
 - Capabilities: Glob, grep, AST queries to locate files, symbols, patterns
 - **Delegate when:** Need to discover what exists before planning • Parallel searches speed discovery • Need summarized map vs full contents • Broad/uncertain scope
 - **Don't delegate when:** Know the path and need actual content • Need full file anyway • Single specific lookup • About to edit the file
 
 @librarian
 - Role: Authoritative source for current library docs and API references
+- Stats: 10x better finding up-to-date library docs than orchestrator, 1/2 cost of orchestrator
 - Capabilities: Fetches latest official docs, examples, API signatures, version-specific behavior via grep_app MCP
 - **Delegate when:** Libraries with frequent API changes (React, Next.js, AI SDKs) • Complex APIs needing official examples (ORMs, auth) • Version-specific behavior matters • Unfamiliar library • Edge cases or advanced features • Nuanced best practices
 - **Don't delegate when:** Standard usage you're confident about (\`Array.map()\`, \`fetch()\`) • Simple stable APIs • General programming knowledge • Info already in conversation • Built-in language features
@@ -44,6 +46,7 @@ You are an AI coding orchestrator that optimizes for quality, speed, cost, and r
 
 @oracle
 - Role: Strategic advisor for high-stakes decisions and persistent problems
+- Stats: 10x better decisions maker, problem solver, investigator than orchestrator, 0.8x speed of orchestrator, same cost.
 - Capabilities: Deep architectural reasoning, system-level trade-offs, complex debugging
 - Tools/Constraints: Slow, expensive, high-quality—use sparingly when thoroughness beats speed
 - **Delegate when:** Major architectural decisions with long-term impact • Problems persisting after 2+ fix attempts • High-risk multi-system refactors • Costly trade-offs (performance vs maintainability) • Complex debugging with unclear root cause • Security/scalability/data integrity decisions • Genuinely uncertain and cost of wrong choice is high
@@ -52,22 +55,24 @@ You are an AI coding orchestrator that optimizes for quality, speed, cost, and r
 
 @designer
 - Role: UI/UX specialist for intentional, polished experiences
+- Stats: 10x better UI/UX than orchestrator
 - Capabilities: Visual direction, interactions, responsive layouts, design systems with aesthetic intent
 - **Delegate when:** User-facing interfaces needing polish • Responsive layouts • UX-critical components (forms, nav, dashboards) • Visual consistency systems • Animations/micro-interactions • Landing/marketing pages • Refining functional→delightful
 - **Don't delegate when:** Backend/logic with no visual • Quick prototypes where design doesn't matter yet
 - **Rule of thumb:** Users see it and polish matters? → @designer. Headless/functional? → yourself.
 
 @fixer
-- Role: Fast, parallel execution specialist for well-defined tasks
-- Capabilities: Efficient implementation when spec and context are clear
+- Role: Fast execution specialist for well-defined tasks, which empowers orchestrator with parallel, speedy executions
+- Stats: 2x faster code edits, 1/2 cost of orchestrator, 0.8x quality of orchestrator
 - Tools/Constraints: Execution-focused—no research, no architectural decisions
-- **Delegate when:** Clearly specified with known approach • 3+ independent parallel tasks • Straightforward but time-consuming • Solid plan needing execution • Repetitive multi-location changes • Overhead < time saved by parallelization
-- **Don't delegate when:** Needs discovery/research/decisions • Single small change (<20 lines, one file) • Unclear requirements needing iteration • Explaining > doing • Tight integration with your current work • Sequential dependencies
-- **Parallelization:** 3+ independent tasks → spawn multiple @fixers. 1-2 simple tasks → do yourself.
-- **Rule of thumb:** Explaining > doing? → yourself. Can split to parallel streams? → multiple @fixers.
+- **Delegate when:** For implementation work, think and triage first. If the change is non-trivial or multi-file, hand bounded execution to @fixer
+- **Don't delegate when:** Needs discovery/research/decisions • Single small change (<20 lines, one file) • Unclear requirements needing iteration • Explaining to fixer > doing • Tight integration with your current work • Sequential dependencies
+- **Rule of thumb:** Explaining > doing? → yourself. Orchestrator paths selection is vastly improved by Fixer. eg it can reduce overall speed if Orchestrator splits what's usually a single task into multiple subtasks and parallelize it with fixer.
+  Fixer also recudes cost by beeing cheaper than orchestrator for implementation tasks.
 
 @council
 - Role: Multi-LLM consensus engine for high-confidence answers
+- Stats: 3x slower than orchestrator, 3x or more cost of orchestrator
 - Capabilities: Runs multiple models in parallel, synthesizes their responses via a council master
 - **Delegate when:** Critical decisions needing diverse model perspectives • High-stakes architectural choices where consensus reduces risk • Ambiguous problems where multi-model disagreement is informative • Security-sensitive design reviews
 - **Don't delegate when:** Straightforward tasks you're confident about • Speed matters more than confidence • Single-model answer is sufficient • Routine implementation work
@@ -81,19 +86,14 @@ You are an AI coding orchestrator that optimizes for quality, speed, cost, and r
 ## 1. Understand
 Parse request: explicit requirements + implicit needs.
 
-## 2. Path Analysis
+## 2. Path Selection
 Evaluate approach by: quality, speed, cost, reliability.
 Choose the path that optimizes all four.
 
 ## 3. Delegation Check
 **STOP. Review specialists before acting.**
 
-Each specialist delivers 10x results in their domain:
-- @explorer → Parallel discovery when you need to find unknowns, not read knowns
-- @librarian → Complex/evolving APIs where docs prevent errors, not basic usage
-- @oracle → High-stakes decisions where wrong choice is costly, not routine calls
-- @designer → User-facing experiences where polish matters, not internal logic
-- @fixer → Parallel execution of clear specs, not explaining trivial changes
+!!! Review available agents and delegation rules. Decide whether to delegate or do it yourself. !!!
 
 **Delegation efficiency:**
 - Reference paths/lines, don't paste files (\`src/app.ts:42\` not full contents)
@@ -101,21 +101,16 @@ Each specialist delivers 10x results in their domain:
 - Brief user on delegation goal before each call
 - Skip delegation if overhead ≥ doing it yourself
 
-**Fixer parallelization:**
-- 3+ independent tasks? Spawn multiple @fixers simultaneously
-- 1-2 simple tasks? Do it yourself
-- Sequential dependencies? Handle serially or do yourself
-
-## 4. Parallelize
-Can tasks run simultaneously?
+## 4. Split and Parallelize
+Can tasks be splitted to subtasks and run in parallel?
 - Multiple @explorer searches across different domains?
 - @explorer + @librarian research in parallel?
-- Multiple @fixer instances for independent changes?
+- Multiple @fixer instances for independent implementation tasks?
 
 Balance: respect dependencies, avoid parallelizing what must be sequential.
 
 ## 5. Execute
-1. Break complex tasks into todos if needed
+1. Break complex tasks into todos
 2. Fire parallel research/implementation
 3. Delegate to specialists or do it yourself based on step 3
 4. Integrate results
@@ -127,10 +122,6 @@ Balance: respect dependencies, avoid parallelizing what must be sequential.
 - Confirm specialists completed successfully
 - Verify solution meets requirements
 
-## Agent Role Mapping
-When a workflow calls for an **implementer** subagent: dispatch \`@fixer\`. Fixer has enforced constraints (no research, no delegation, structured output) that match the implementer role exactly.
-When a workflow calls for a **reviewer** subagent: dispatch \`@oracle\`. Oracle has the depth for architectural review and access to code review skills.
-
 </Workflow>
 
 <Communication>

src/config/constants.ts

@@ -78,6 +78,11 @@ export const FALLBACK_FAILOVER_TIMEOUT_MS = 15_000;
 // Subagent depth limits
 export const DEFAULT_MAX_SUBAGENT_DEPTH = 3;
 
+// Workflow reminders
+export const PHASE_REMINDER_TEXT = `Recall Workflow Rules:
+Understand → build the best path (delegated based on rules, splitted and parallelized as much as possible) → execute → verify.
+If delegating, launch the specialist in the same turn you mention it.`;
+
 // Tmux pane spawn delay (ms) — gives TmuxSessionManager time to create pane
 export const TMUX_SPAWN_DELAY_MS = 500;
 

src/hooks/index.ts

@@ -8,4 +8,4 @@ export {
 } from './foreground-fallback';
 export { createJsonErrorRecoveryHook } from './json-error-recovery';
 export { createPhaseReminderHook } from './phase-reminder';
-export { createPostReadNudgeHook } from './post-read-nudge';
+export { createPostFileToolNudgeHook } from './post-file-tool-nudge';

src/hooks/phase-reminder/index.ts

@@ -8,11 +8,10 @@
  *
  * Uses experimental.chat.messages.transform so it doesn't show in UI.
  */
+import { PHASE_REMINDER_TEXT } from '../../config/constants';
 import { SLIM_INTERNAL_INITIATOR_MARKER } from '../../utils';
 
-export const PHASE_REMINDER = `<reminder>Recall Workflow Rules:
-Understand → find the best path (delegate based on rules and parallelize independent work) → execute → verify.
-If delegating, launch the specialist in the same turn you mention it.</reminder>`;
+export const PHASE_REMINDER = `<reminder>${PHASE_REMINDER_TEXT}</reminder>`;
 
 interface MessageInfo {
   role: string;

src/hooks/post-file-tool-nudge/codemap.md

@@ -0,0 +1,19 @@
+# src/hooks/post-file-tool-nudge/
+
+<!-- Explorer: Fill in this section with architectural understanding -->
+
+## Responsibility
+
+Provide a lightweight safety net that reminds agents to stay in the delegation workflow whenever they read or write project files. When the Read or Write tool emits output, this hook appends a standardized nudge so that follow-up work is routed to the right specialists instead of being solved immediately.
+
+## Design
+
+Exports a single factory (`createPostFileToolNudgeHook`) that returns a handler map for the `tool.execute.after` event. The hook keeps a shared `NUDGE` constant and enforces a policy guard for Read/Write tool names before mutating the `output`. This keeps the implementation focused and pluggable inside the broader hook registry.
+
+## Flow
+
+The hook is instantiated once and registered with the hook system, which invokes `tool.execute.after` after every tool call. When a Read or Write tool completes, the hook sees the input metadata, verifies the tool name, and appends the reminder text to `output.output`, thereby altering the user-facing response before it bubbles back to the agent.
+
+## Integration
+
+Plugged into the global hook registry, this module intercepts every tool response via the `tool.execute.after` lifecycle event. It directly touches only Read/Write tool output objects so there are no downstream dependencies, but the appended reminder influences all downstream consumers who display file-tool results.

src/hooks/post-file-tool-nudge/index.ts

@@ -0,0 +1,42 @@
+/**
+ * Post-tool nudge - appends a delegation reminder after file reads/writes.
+ * Catches the "inspect/edit files → implement myself" anti-pattern.
+ */
+
+import { PHASE_REMINDER_TEXT } from '../../config/constants';
+
+const NUDGE = `\n\n---\n${PHASE_REMINDER_TEXT}`;
+
+interface ToolExecuteAfterInput {
+  tool: string;
+  sessionID?: string;
+  callID?: string;
+}
+
+interface ToolExecuteAfterOutput {
+  title: string;
+  output: string;
+  metadata: Record<string, unknown>;
+}
+
+export function createPostFileToolNudgeHook() {
+  return {
+    'tool.execute.after': async (
+      input: ToolExecuteAfterInput,
+      output: ToolExecuteAfterOutput,
+    ): Promise<void> => {
+      // Only nudge for Read/Write tools
+      if (
+        input.tool !== 'Read' &&
+        input.tool !== 'read' &&
+        input.tool !== 'Write' &&
+        input.tool !== 'write'
+      ) {
+        return;
+      }
+
+      // Append the nudge
+      output.output = output.output + NUDGE;
+    },
+  };
+}

src/hooks/post-read-nudge/codemap.md

@@ -1,19 +0,0 @@
-# src/hooks/post-read-nudge/
-
-<!-- Explorer: Fill in this section with architectural understanding -->
-
-## Responsibility
-
-Provide a lightweight safety net that reminds agents to stay in the delegation workflow whenever they read project files. When the Read tool emits output, this hook appends a standardized nudge so that follow-up work is routed to the right specialists instead of being solved immediately.
-
-## Design
-
-Exports a single factory (`createPostReadNudgeHook`) that returns a handler map for the `tool.execute.after` event. The hook keeps a shared `NUDGE` constant and enforces a policy guard (`input.tool === 'Read' || input.tool === 'read'`) before mutating the `output`. This keeps the implementation focused and pluggable inside the broader hook registry.
-
-## Flow
-
-The hook is instantiated once and registered with the hook system, which invokes `tool.execute.after` after every tool call. When the Read tool completes, the hook sees the input metadata, verifies the tool name, and appends the reminder text to `output.output`, thereby altering the user-facing response before it bubbles back to the agent.
-
-## Integration
-
-Plugged into the global hook registry, this module intercepts every tool response via the `tool.execute.after` lifecycle event. It directly touches only the Read tool’s output object so there are no downstream dependencies, but the appended reminder influences all downstream consumers who display read results.

src/hooks/post-read-nudge/index.ts

@@ -1,36 +0,0 @@
-/**
- * Post-Read nudge - appends a delegation reminder after file reads.
- * Catches the "read files → implement myself" anti-pattern.
- */
-
-const NUDGE =
-  '\n\n---\nWorkflow Reminder: delegate based on rules; If mentioning a specialist, launch it in this same turn.';
-
-interface ToolExecuteAfterInput {
-  tool: string;
-  sessionID?: string;
-  callID?: string;
-}
-
-interface ToolExecuteAfterOutput {
-  title: string;
-  output: string;
-  metadata: Record<string, unknown>;
-}
-
-export function createPostReadNudgeHook() {
-  return {
-    'tool.execute.after': async (
-      input: ToolExecuteAfterInput,
-      output: ToolExecuteAfterOutput,
-    ): Promise<void> => {
-      // Only nudge for Read tool
-      if (input.tool !== 'Read' && input.tool !== 'read') {
-        return;
-      }
-
-      // Append the nudge
-      output.output = output.output + NUDGE;
-    },
-  };
-}

src/index.ts

@@ -10,7 +10,7 @@ import {
   createDelegateTaskRetryHook,
   createJsonErrorRecoveryHook,
   createPhaseReminderHook,
-  createPostReadNudgeHook,
+  createPostFileToolNudgeHook,
   ForegroundFallbackManager,
 } from './hooks';
 import { createBuiltinMcps } from './mcp';
@@ -124,8 +124,8 @@ const OhMyOpenCodeLite: Plugin = async (ctx) => {
   // Initialize phase reminder hook for workflow compliance
   const phaseReminderHook = createPhaseReminderHook();
 
-  // Initialize post-read nudge hook
-  const postReadNudgeHook = createPostReadNudgeHook();
+  // Initialize post-file-tool nudge hook
+  const postFileToolNudgeHook = createPostFileToolNudgeHook();
 
   const chatHeadersHook = createChatHeadersHook(ctx);
 
@@ -387,7 +387,7 @@ const OhMyOpenCodeLite: Plugin = async (ctx) => {
     'experimental.chat.messages.transform':
       phaseReminderHook['experimental.chat.messages.transform'],
 
-    // Post-tool hooks: retry guidance for delegation errors + post-read nudge
+    // Post-tool hooks: retry guidance for delegation errors + file-tool nudge
     'tool.execute.after': async (input, output) => {
       await delegateTaskRetryHook['tool.execute.after'](
         input as { tool: string },
@@ -407,7 +407,7 @@ const OhMyOpenCodeLite: Plugin = async (ctx) => {
         },
       );
 
-      await postReadNudgeHook['tool.execute.after'](
+      await postFileToolNudgeHook['tool.execute.after'](
         input as {
           tool: string;
           sessionID?: string;