Connected machine result

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/GoAnyWhere/AGENTS.md`
- Agent type: Universal agent guidance
- Injection mode: Append or create. Most coding agents read this as folder-local guidance.

### Scope
Apply this policy to: `/Users/macos/Desktop/GoAnyWhere`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/GoAnyWhere/CLAUDE.md`
- Agent type: Claude guidance
- Injection mode: Append Claude-specific model routing, context, and loop rules.

### Scope
Apply this policy to: `/Users/macos/Desktop/GoAnyWhere`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/GoAnyWhere/GEMINI.md`
- Agent type: Gemini guidance
- Injection mode: Append Gemini-specific low-cost exploration and synthesis rules.

### Scope
Apply this policy to: `/Users/macos/Desktop/GoAnyWhere`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

---
description: AgenticGuide efficiency policy for this workspace area
globs:
  - "**/*"
alwaysApply: false
---

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/GoAnyWhere/.cursor/rules/agentic-guide.mdc`
- Agent type: Cursor rule
- Injection mode: Create as a Cursor project rule for targeted agent behavior.

### Scope
Apply this policy to: `/Users/macos/Desktop/GoAnyWhere`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/personal_project/AGENTS.md`
- Agent type: Universal agent guidance
- Injection mode: Append or create. Most coding agents read this as folder-local guidance.

### Scope
Apply this policy to: `/Users/macos/Desktop/personal_project`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/personal_project/CLAUDE.md`
- Agent type: Claude guidance
- Injection mode: Append Claude-specific model routing, context, and loop rules.

### Scope
Apply this policy to: `/Users/macos/Desktop/personal_project`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/personal_project/GEMINI.md`
- Agent type: Gemini guidance
- Injection mode: Append Gemini-specific low-cost exploration and synthesis rules.

### Scope
Apply this policy to: `/Users/macos/Desktop/personal_project`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

---
description: AgenticGuide efficiency policy for this workspace area
globs:
  - "**/*"
alwaysApply: false
---

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/personal_project/.cursor/rules/agentic-guide.mdc`
- Agent type: Cursor rule
- Injection mode: Create as a Cursor project rule for targeted agent behavior.

### Scope
Apply this policy to: `/Users/macos/Desktop/personal_project`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/SalixProj/AGENTS.md`
- Agent type: Universal agent guidance
- Injection mode: Append or create. Most coding agents read this as folder-local guidance.

### Scope
Apply this policy to: `/Users/macos/Desktop/SalixProj`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/SalixProj/CLAUDE.md`
- Agent type: Claude guidance
- Injection mode: Append Claude-specific model routing, context, and loop rules.

### Scope
Apply this policy to: `/Users/macos/Desktop/SalixProj`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/SalixProj/GEMINI.md`
- Agent type: Gemini guidance
- Injection mode: Append Gemini-specific low-cost exploration and synthesis rules.

### Scope
Apply this policy to: `/Users/macos/Desktop/SalixProj`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

---
description: AgenticGuide efficiency policy for this workspace area
globs:
  - "**/*"
alwaysApply: false
---

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/SalixProj/.cursor/rules/agentic-guide.mdc`
- Agent type: Cursor rule
- Injection mode: Create as a Cursor project rule for targeted agent behavior.

### Scope
Apply this policy to: `/Users/macos/Desktop/SalixProj`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/AGENTS.md`
- Agent type: Universal agent guidance
- Injection mode: Append or create. Most coding agents read this as folder-local guidance.

### Scope
Apply this policy to: `/Users/macos/Desktop`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/CLAUDE.md`
- Agent type: Claude guidance
- Injection mode: Append Claude-specific model routing, context, and loop rules.

### Scope
Apply this policy to: `/Users/macos/Desktop`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/GEMINI.md`
- Agent type: Gemini guidance
- Injection mode: Append Gemini-specific low-cost exploration and synthesis rules.

### Scope
Apply this policy to: `/Users/macos/Desktop`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

---
description: AgenticGuide efficiency policy for this workspace area
globs:
  - "**/*"
alwaysApply: false
---

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/.cursor/rules/agentic-guide.mdc`
- Agent type: Cursor rule
- Injection mode: Create as a Cursor project rule for targeted agent behavior.

### Scope
Apply this policy to: `/Users/macos/Desktop`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/personal_project/Ruai_s/AGENTS.md`
- Agent type: Universal agent guidance
- Injection mode: Append or create. Most coding agents read this as folder-local guidance.

### Scope
Apply this policy to: `/Users/macos/Desktop/personal_project/Ruai_s`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/personal_project/Ruai_s/CLAUDE.md`
- Agent type: Claude guidance
- Injection mode: Append Claude-specific model routing, context, and loop rules.

### Scope
Apply this policy to: `/Users/macos/Desktop/personal_project/Ruai_s`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/personal_project/Ruai_s/GEMINI.md`
- Agent type: Gemini guidance
- Injection mode: Append Gemini-specific low-cost exploration and synthesis rules.

### Scope
Apply this policy to: `/Users/macos/Desktop/personal_project/Ruai_s`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.

---
description: AgenticGuide efficiency policy for this workspace area
globs:
  - "**/*"
alwaysApply: false
---

## Agentic-Guide Optimization Policy

### Target
- Agent file: `/Users/macos/Desktop/personal_project/Ruai_s/.cursor/rules/agentic-guide.mdc`
- Agent type: Cursor rule
- Injection mode: Create as a Cursor project rule for targeted agent behavior.

### Scope
Apply this policy to: `/Users/macos/Desktop/personal_project/Ruai_s`

### Detected Issues
- Expensive model overuse (high, 82% confidence): 1623 premium model calls out of 1683
- Search loop (medium, 74% confidence): grep repeated 159 times; find repeated 20 times
- Command repetition (medium, 76% confidence): head ran 222 times; echo ran 181 times
- High context usage (high, 80% confidence): GPT-5.5 used 525240 tokens; GPT-5.5 used 525240 tokens

### Model Usage Rules
- Use premium reasoning models only for architecture decisions, security review, and final synthesis.
- Use faster economical models for search, file inventory, formatting, and low-risk summarization.
- Escalate model strength only after writing the reason in the working notes.

### Context Rules
- Read the smallest relevant file set first and summarize findings before expanding scope.
- Reuse prior observations instead of reopening the same files repeatedly.
- Avoid loading generated folders, dependency folders, secrets, and build output into context.

### Loop Prevention
- Route exploration and file lookup work to economical models; reserve premium models for planning, review, and final synthesis.
- Record search intent and results before issuing another equivalent search.
- Add stop conditions for repeated commands and summarize command output before retrying.
- Use file-scoped retrieval and compact summaries before expanding context.
- Stop after three equivalent searches or commands and produce a decision checkpoint.

### Editing Rules
- State the intended write set before editing this folder.
- Keep changes scoped to the current objective and verify behavior after edits.
- Do not modify secrets, generated assets, lockfiles, or migration history without explicit approval.

### Expected Impact
- Lower model spend from reduced premium-model exploration.
- Fewer repeated searches and file reads.
- Safer folder-local agent behavior with clearer edit boundaries.