cokit-cli 1.0.4 → 1.0.5

package/templates/repo/.github/prompts/ck-skill-fix-logs.prompt.md

@@ -0,0 +1,22 @@
+---
+agent: 'agent'
+description: 'Fix the agent skill based on `logs.txt` file.'
+tools: ['search/codebase', 'search/changes', 'web/fetch', 'read/problems', 'read/terminalLastCommand']
+---
+
+Think harder.
+Use `skill-creator` and `claude-code` skills.
+
+## Your mission
+Fix the agent skill based on the current `logs.txt` file (in the project root directory).
+
+## Requirements
+<user-prompt>${input}</user-prompt>
+
+## Rules of Skill Fixing:
+Base on the requirements:
+- If you're given nothing, use `AskUserQuestion` tool for clarifications and  to research about the topic.
+- If you're given an URL, it's documentation page, explore every internal link and .
+- If you receive a lot of URLs, use multiple s to explore them , .
+- If you receive a lot of files, use multiple s to explore them , .
+- If you're given a Github URL, use [](https://repomix.com/guide/usage) command to summarize ([install it](https://repomix.com/guide/installation) if needed) and spawn multiple s to explore it , .

package/templates/repo/.github/prompts/ck-skill-optimize-auto.prompt.md

@@ -0,0 +1,25 @@
+---
+agent: 'agent'
+description: 'Optimize an existing agent skill [auto]'
+tools: ['search/codebase', 'search/changes', 'web/fetch', 'read/problems']
+---
+
+Think harder.
+Use `skill-creator` and `claude-code` skills.
+
+## Arguments
+SKILL: $1 (default: `*`)
+PROMPT: $2 (default: empty)
+
+## Your mission
+Optimize an existing skill in  directory. 
+Always keep in mind that `SKILL.md` and reference files should be token consumption efficient, so that **progressive disclosure** can be leveraged at best.
+`SKILL.md` is always short and concise, straight to the point, treat it as a quick reference guide.
+
+**IMPORTANT:**
+- Skills are not documentation, they are practical instructions for Claude Code to use the tools, packages, plugins or APIs to achieve the tasks.
+- Each skill teaches Claude how to perform a specific development task, not what a tool does.
+- Claude Code can activate multiple skills automatically to achieve the user's request.
+
+## Additional instructions
+<additional-instructions>$PROMPT</additional-instructions>

package/templates/repo/.github/prompts/ck-skill-optimize.prompt.md

@@ -0,0 +1,34 @@
+---
+agent: 'agent'
+description: 'Optimize an existing agent skill'
+tools: ['search/codebase', 'search/changes', 'web/fetch', 'read/problems']
+---
+
+Think harder.
+Use `skill-creator` and `claude-code` skills.
+
+## Arguments
+SKILL: $1 (default: `*`)
+PROMPT: $2 (default: empty)
+
+## Your mission
+Propose a plan to optimize an existing skill in  directory. 
+When you finish, ask user to review your plan:
+- If the user approve: Write down a plan follow "Output Requirements", then ask user if they want to start implementing.
+- If the user reject: Revise the plan or ask more questions to clarify more about the user's request (ask one question at the time), then repeat the review process.
+
+## Additional instructions
+<additional-instructions>$PROMPT</additional-instructions>
+
+## Output Requirements
+An output implementation plan must also follow the progressive disclosure structure:
+- Always keep in mind that `SKILL.md` and reference files should be token consumption efficient, so that **progressive disclosure** can be leveraged at best.
+- `SKILL.md` is always short and concise, straight to the point, treat it as a quick reference guide.
+- Create a directory using naming pattern from `## Naming` section.
+- Save the overview access point at `plan.md`, keep it generic, under 80 lines, and list each phase with status/progress and links.
+- For each phase, add `phase-XX-phase-name.md` files containing sections (Context links, Overview with date/priority/statuses, Key Insights, Requirements, Architecture, Related code files, Implementation Steps, Todo list, Success Criteria, Risk Assessment, Security Considerations, Next steps).
+
+**IMPORTANT:**
+- Skills are not documentation, they are practical instructions for Claude Code to use the tools, packages, plugins or APIs to achieve the tasks.
+- Each skill teaches Claude how to perform a specific development task, not what a tool does.
+- Claude Code can activate multiple skills automatically to achieve the user's request.

package/templates/repo/.github/prompts/ck-skill-plan.prompt.md

@@ -0,0 +1,46 @@
+---
+agent: 'agent'
+description: 'Plan to create a new agent skill'
+tools: ['search/codebase', 'search/changes', 'web/fetch', 'read/problems']
+---
+
+Think harder.
+First: Activate `skill-creator` and `claude-code` skills.
+
+Employ `sequential-thinking` or `problem-solving` skills for complex problem-solving that requires structured analysis
+Finally, when creating a plan, 
+
+## Arguments
+SKILL: $1 (default: `*`)
+PROMPT: $2 (default: empty)
+
+## Your mission
+Propose a plan to create a new skill in  directory. 
+When you finish, ask user to review your plan:
+- If the user approve: Write down a plan follow "Output Requirements", then ask user if they want to start implementing.
+- If the user reject: Revise the plan or ask more questions to clarify more about the user's request (ask one question at the time), then repeat the review process.
+
+## Additional instructions
+<additional-instructions>$PROMPT</additional-instructions>
+
+## Your Approach
+1. **Question Everything**: Ask probing questions to fully understand the user's request, constraints, and true objectives. Don't assume - clarify until you're 100% certain.
+2. **Brutal Honesty**: Provide frank, unfiltered feedback about ideas. If something is unrealistic, over-engineered, or likely to cause problems, say so directly. Your job is to prevent costly mistakes.
+3. **Explore Alternatives**: Always consider multiple approaches. Present 2-3 viable solutions with clear pros/cons, explaining why one might be superior.
+4. **Challenge Assumptions**: Question the user's initial approach. Often the best solution is different from what was originally envisioned.
+5. **Consider All Stakeholders**: Evaluate impact on end users, developers, operations team, and business objectives.
+
+## Output Requirements
+An output implementation plan must also follow the progressive disclosure structure:
+- Always keep in mind that `SKILL.md` and reference files should be token consumption efficient, so that **progressive disclosure** can be leveraged at best.
+- `SKILL.md` is always short and concise, straight to the point, treat it as a quick reference guide.
+- Create a directory using naming pattern from `## Naming` section.
+- Save the overview access point at `plan.md`, keep it generic, under 80 lines, and list each phase with status/progress and links.
+- For each phase, add `phase-XX-phase-name.md` files containing sections (Context links, Overview with date/priority/statuses, Key Insights, Requirements, Architecture, Related code files, Implementation Steps, Todo list, Success Criteria, Risk Assessment, Security Considerations, Next steps).
+
+**IMPORTANT:**
+- Analyze the given task and use `AskUserQuestion` tool to ask for more details if needed.
+- Ensure token consumption efficiency while maintaining high quality.
+- Skills are not documentation, they are practical instructions for Claude Code to use the tools, packages, plugins or APIs to achieve the tasks.
+- Each skill teaches Claude how to perform a specific development task, not what a tool does.
+- Claude Code can activate multiple skills automatically to achieve the user's request.

package/templates/repo/.github/prompts/ck-test-ui.prompt.md

@@ -0,0 +1,92 @@
+---
+agent: 'agent'
+description: 'Run UI tests on a website & generate a detailed report.'
+tools: ['search/codebase', 'search/changes', 'web/fetch', 'read/problems', 'read/terminalLastCommand']
+---
+
+Activate the chrome-devtools skill.
+
+## Purpose
+Run comprehensive UI tests on a website and generate a detailed report.
+
+## Arguments
+- $1: URL - The URL of the website to test
+- $2: OPTIONS - Optional test configuration (e.g., --headless, --mobile, --auth)
+
+## Testing Protected Routes (Authentication)
+
+For testing protected routes that require authentication, follow this workflow:
+
+### Step 1: User Manual Login
+Instruct the user to:
+1. Open the target site in their browser
+2. Log in manually with their credentials
+3. Open browser DevTools (F12) → Application tab → Cookies/Storage
+
+### Step 2: Extract Auth Credentials
+Ask the user to provide one of:
+- **Cookies**: Copy cookie values (name, value, domain)
+- **Access Token**: Copy JWT/Bearer token from localStorage or cookies
+- **Session Storage**: Copy relevant session keys
+
+### Step 3: Inject Authentication
+Use the `inject-auth.js` script to inject credentials before testing:
+
+```bash
+cd $SKILL_DIR  # 
+
+# Option A: Inject cookies
+node inject-auth.js --url https://example.com --cookies '[{"name":"session","value":"abc123","domain":".example.com"}]'
+
+# Option B: Inject Bearer token
+node inject-auth.js --url https://example.com --token "Bearer eyJhbGciOi..." --header Authorization --token-key access_token
+
+# Option C: Inject localStorage
+node inject-auth.js --url https://example.com --local-storage '{"auth_token":"xyz","user_id":"123"}'
+
+# Combined (cookies + localStorage)
+node inject-auth.js --url https://example.com --cookies '[{"name":"session","value":"abc"}]' --local-storage '{"user":"data"}'
+```
+
+### Step 4: Run Tests
+After auth injection, the browser session persists. Run tests normally:
+
+```bash
+# Navigate and screenshot protected pages
+node navigate.js --url https://example.com/dashboard
+node screenshot.js --url https://example.com/profile --output profile.png
+
+# The auth session persists until --close true is used
+node screenshot.js --url https://example.com/settings --output settings.png --close true
+```
+
+### Auth Script Options
+- `--cookies '<json>'` - Inject cookies (JSON array)
+- `--token '<token>'` - Inject Bearer token
+- `--token-key '<key>'` - localStorage key for token (default: access_token)
+- `--header '<name>'` - Set HTTP header with token (e.g., Authorization)
+- `--local-storage '<json>'` - Inject localStorage items
+- `--session-storage '<json>'` - Inject sessionStorage items
+- `--reload true` - Reload page after injection
+- `--clear true` - Clear saved auth session
+
+## Workflow
+- All the screenshots should be saved in the same report directory.
+- Browse $URL with the specified $OPTIONS, discover all pages, components, and endpoints.
+- Create a test plan based on the discovered structure
+- Use multiple s or tool calls  to test all pages, forms, navigation, user flows, accessibility, functionalities, usability, responsive layouts, cross-browser compatibility, performance, security, seo, etc.
+- Use `ai-multimodal` to analyze all screenshots and visual elements.
+- Generate a comprehensive report in Markdown format, embedding all screenshots directly in the report.
+- Finally respond to the user with a concise summary of findings and recommendations.
+- Use `AskUserQuestion` tool to ask if user wants to preview the report with `/preview` .
+
+## Output Requirements
+How to write reports:
+- Format: Use clear, structured Markdown with headers, lists, and code blocks where appropriate
+- Include the test results summary, key findings, and screenshot references
+- **IMPORTANT:** Ensure token efficiency while maintaining high quality.
+- **IMPORTANT:** Sacrifice grammar for the sake of concision when writing reports.
+- **IMPORTANT:** In reports, list any unresolved questions at the end, if any.
+
+**IMPORTANT**: **Do not** start implementing the fixes.
+**IMPORTANT:** 

package/templates/repo/.github/prompts/ck-test.prompt.md

@@ -1,23 +1,10 @@
 ---
-mode: agent
-description: Write and run tests
+agent: 'agent'
+description: 'Run tests locally and analyze the summary report.'
+tools: ['search/codebase', 'search/changes', 'read/problems', 'read/terminalLastCommand']
 ---
 
-# Test - Testing
+run tests locally and analyze the summary report.
 
-You are a testing expert. Help ensure code quality through testing.
-
-## Approach
-
-1. **Understand what to test** - What's the expected behavior?
-2. **Write tests** - Cover happy path, edge cases, errors
-3. **Run tests** - Execute and check results
-4. **Fix failures** - Debug and fix failing tests
-
-## Test Guidelines
-
-- Test behavior, not implementation
-- Use descriptive test names
-- Keep tests focused and independent
-- Mock external dependencies
-- Cover edge cases
+**IMPORTANT**: **Do not** start implementing.
+**IMPORTANT:** 

package/templates/repo/.github/prompts/ck-use-mcp.prompt.md

@@ -0,0 +1,27 @@
+---
+agent: 'agent'
+description: 'Utilize tools of Model Context Protocol (MCP) servers'
+tools: ['search/codebase', 'read/terminalLastCommand']
+---
+
+Execute MCP operations via **Gemini CLI** to preserve context budget.
+
+## Execution Steps
+
+1. **Execute task via Gemini CLI** (using stdin pipe for MCP support):
+   ```bash
+   # IMPORTANT: Use stdin piping, NOT -p flag (deprecated, skips MCP init)
+   echo "${input}. Return JSON only per GEMINI.md instructions." | gemini -y -m gemini-2.5-flash
+   ```
+
+2. **Fallback to mcp-manager ** (if Gemini CLI unavailable):
+   - discover and execute tools
+   - If the  got issues with the scripts of `mcp-management` skill, md auto-loaded**: Gemini CLI automatically loads `GEMINI.md` from project root, enforcing JSON-only response format
+- **Parseable output**: Responses are structured JSON: `{"server":"name","tool":"name","success":true,"result":<data>,"error":null}`
+
+## Anti-Pattern (DO NOT USE)
+
+```bash
+# BROKEN - deprecated -p flag skips MCP server connections!
+gemini -y -m gemini-2.5-flash -p "..."
+```

package/templates/repo/.github/prompts/ck-watzup.prompt.md

@@ -0,0 +1,11 @@
+---
+agent: 'agent'
+description: 'Review recent changes and wrap up the work'
+tools: ['search/codebase', 'search/changes', 'read/problems']
+---
+
+Review my current branch and the most recent commits. 
+Provide a detailed summary of all changes, including what was modified, added, or removed. 
+Analyze the overall impact and quality of the changes.
+
+**IMPORTANT**: **Do not** start implementing.

package/templates/repo/.github/prompts/ck-worktree.prompt.md

@@ -0,0 +1,127 @@
+---
+agent: 'agent'
+description: 'Create isolated git worktree for parallel development'
+tools: ['search/codebase', 'read/terminalLastCommand']
+---
+
+Create an isolated git worktree for parallel feature development.
+
+## Workflow
+
+### Step 1: Get Repository Info
+
+```bash
+node  info --json
+```
+
+**Response fields:**
+- `repoType`: "monorepo" or "standalone"
+- `baseBranch`: detected base branch
+- `projects`: array of {name, path} for monorepo
+- `envFiles`: array of .env* files found
+- `dirtyState`: boolean
+
+### Step 2: Gather Info via AskUserQuestion
+
+**Detect branch prefix from user's description:**
+- Keywords "fix", "bug", "error", "issue" → prefix = `fix`
+- Keywords "refactor", "restructure", "rewrite" → prefix = `refactor`
+- Keywords "docs", "documentation", "readme" → prefix = `docs`
+- Keywords "test", "spec", "coverage" → prefix = `test`
+- Keywords "chore", "cleanup", "deps" → prefix = `chore`
+- Keywords "perf", "performance", "optimize" → prefix = `perf`
+- Everything else → prefix = `feat`
+
+**For MONOREPO:** Use AskUserQuestion if project not specified:
+```javascript
+// If user said "/worktree add auth" but multiple projects exist
+AskUserQuestion({
+  questions: [{
+    header: "Project",
+    question: "Which project should the worktree be created for?",
+    options: projects.map(p => ({ label: p.name, description: p.path })),
+    multiSelect: false
+  }]
+})
+```
+
+**For env files:** Always ask which to copy:
+```javascript
+AskUserQuestion({
+  questions: [{
+    header: "Env files",
+    question: "Which environment files should be copied to the worktree?",
+    options: envFiles.map(f => ({ label: f, description: "Copy to worktree" })),
+    multiSelect: true
+  }]
+})
+```
+
+### Step 3: Convert Description to Slug
+
+- "add authentication system" → `add-auth`
+- "fix login bug" → `login-bug`
+- Remove filler words, kebab-case, max 50 chars
+
+### Step 4: Execute Command
+
+**Monorepo:**
+```bash
+node  create "<PROJECT>" "<SLUG>" --prefix <TYPE> --env "<FILES>"
+```
+
+**Standalone:**
+```bash
+node  create "<SLUG>" --prefix <TYPE> --env "<FILES>"
+```
+
+**Options:**
+- `--prefix` - Branch type: feat|fix|refactor|docs|test|chore|perf
+- `--env` - Comma-separated .env files to copy
+- `--json` - Output JSON for parsing
+- `--dry-run` - Preview without executing
+
+## Commands
+
+| Command | Usage | Description |
+|---------|-------|-------------|
+| `create` | `create [project] <feature>` | Create new worktree |
+| `remove` | `remove <name-or-path>` | Remove worktree and branch |
+| `info` | `info` | Get repo info |
+| `list` | `list` | List existing worktrees |
+
+## Error Codes
+
+| Code | Meaning | Action |
+|------|---------|--------|
+| `MISSING_ARGS` | Missing project/feature for monorepo | Ask for both |
+| `MISSING_FEATURE` | No feature name (standalone) | Ask for feature |
+| `PROJECT_NOT_FOUND` | Project not in .gitmodules | Show available projects |
+| `MULTIPLE_PROJECTS_MATCH` | Ambiguous project name | Use AskUserQuestion |
+| `MULTIPLE_WORKTREES_MATCH` | Ambiguous worktree for remove | Use AskUserQuestion |
+| `BRANCH_CHECKED_OUT` | Branch in use elsewhere | Suggest different name |
+| `WORKTREE_EXISTS` | Path already exists | Suggest use or remove |
+| `WORKTREE_CREATE_FAILED` | Git command failed | Show git error |
+| `WORKTREE_REMOVE_FAILED` | Cannot remove worktree | Check uncommitted changes |
+
+## Example Session
+
+```
+User: /worktree fix the login validation bug
+
+Claude: [Runs: node  info --json]
+        [Detects: standalone repo, envFiles: [".env.example"]]
+        [Detects prefix from "fix" keyword: fix]
+        [Converts slug: "login-validation-bug"]
+
+Claude: [Uses AskUserQuestion for env files]
+        "Which environment files should be copied?"
+        Options: .env.example
+
+User: .env.example
+
+Claude: [Runs: node  create "login-validation-bug" --prefix fix --env ".env.example"]
+
+Output: Worktree created at ../worktrees/myrepo-login-validation-bug
+        Branch: fix/login-validation-bug
+```

package/templates/repo/.github/skills/ck-ai-artist/SKILL.md

@@ -0,0 +1,73 @@
+---
+name: 'ck-ai-artist'
+description: 'Write and optimize prompts for AI-generated outcomes across text and image models. Use when crafting prompts for LLMs (Claude, GPT, Gemini), image generators (Midjourney, DALL-E, Stable Diffusion, Imagen, Flux), or video generators (Veo, Runway). Covers prompt structure, style keywords, negative prompts, chain-of-thought, few-shot examples, iterative refinement, and domain-specific patterns for marketing, code, and creative writing.'
+---
+
+# AI Artist - Prompt Engineering
+
+Craft effective prompts for AI text and image generation models.
+
+## Core Principles
+
+1. **Clarity** - Be specific, avoid ambiguity
+2. **Context** - Set scene, role, constraints upfront
+3. **Structure** - Use consistent formatting (markdown, XML tags, delimiters)
+4. **Iteration** - Refine based on outputs, A/B test variations
+
+## Quick Patterns
+
+### LLM Prompts (Claude/GPT/Gemini)
+
+```
+[Role] You are a {expert type} specializing in {domain}.
+[Context] {Background information and constraints}
+[Task] {Specific action to perform}
+[Format] {Output structure - JSON, markdown, list, etc.}
+[Examples] {1-3 few-shot examples if needed}
+```
+
+### Image Generation (Midjourney/DALL-E/Stable Diffusion)
+
+```
+[Subject] {main subject with details}
+[Style] {artistic style, medium, artist reference}
+[Composition] {framing, angle, lighting}
+[Quality] {resolution modifiers, rendering quality}
+[Negative] {what to avoid - only if supported}
+```
+
+**Example**: `Portrait of a cyberpunk hacker, neon lighting, cinematic composition, detailed face, 8k, artstation quality --ar 16:9 --style raw`
+
+## References
+
+Load for detailed guidance:
+
+| Topic | File | Description |
+|-------|------|-------------|
+| LLM | `references/llm-prompting.md` | System prompts, few-shot, CoT, output formatting |
+| Image | `references/image-prompting.md` | Style keywords, model syntax, negative prompts |
+| Nano Banana | `references/nano-banana.md` | Gemini image prompting, narrative style, multi-image input |
+| Advanced | `references/advanced-techniques.md` | Meta-prompting, chaining, A/B testing |
+| Domain Index | `references/domain-patterns.md` | Universal pattern, links to domain files |
+| Marketing | `references/domain-marketing.md` | Headlines, product copy, emails, ads |
+| Code | `references/domain-code.md` | Functions, review, refactoring, debugging |
+| Writing | `references/domain-writing.md` | Stories, characters, dialogue, editing |
+| Data | `references/domain-data.md` | Extraction, analysis, comparison |
+
+## Model-Specific Tips
+
+| Model | Key Syntax |
+|-------|------------|
+| Midjourney | `--ar`, `--style`, `--chaos`, `--weird`, `--v 6.1` |
+| DALL-E 3 | Natural language, no parameters, HD quality option |
+| Stable Diffusion | Weighted tokens `(word:1.2)`, LoRA, negative prompt |
+| Flux | Natural prompts, style mixing, `--guidance` |
+| Imagen/Veo | Descriptive text, aspect ratio, style references |
+
+## Anti-Patterns
+
+- Vague instructions ("make it better")
+- Conflicting constraints
+- Missing context for domain tasks
+- Over-prompting with redundant details
+- Ignoring model-specific strengths/limits

package/templates/repo/.github/skills/ck-ai-artist/references/advanced-techniques.md

@@ -0,0 +1,184 @@
+# Advanced Prompt Engineering
+
+## Prompt Optimization
+
+### DSPy Framework
+Automatic prompt optimization through:
+1. Define task with input/output signatures
+2. Compile with optimizer (BootstrapFewShot, MIPRO)
+3. Model learns optimal prompting strategy
+4. Export optimized prompts for production
+
+### Meta-Prompting
+```
+You are a prompt engineer. Create 5 variations for [task]:
+1. Direct instruction approach
+2. Role-based approach
+3. Few-shot example approach
+4. Chain of thought approach
+5. Constraint-focused approach
+
+Evaluate each, select best.
+```
+
+### Self-Refinement Loop
+```
+Generate: [Initial response]
+Critique: "What's wrong? Score 1-10."
+Refine: "Fix issues, improve score."
+Repeat until score ≥ 8.
+```
+
+## Prompt Chaining
+
+### Sequential Chain
+```
+Chain 1: [Input] → Extract key points
+Chain 2: Key points → Create outline
+Chain 3: Outline → Write draft
+Chain 4: Draft → Edit and polish
+```
+
+### Parallel Chain
+Run independent subtasks simultaneously, merge results.
+
+### Conditional Chain
+```
+If [condition A]: Execute prompt variant 1
+If [condition B]: Execute prompt variant 2
+Else: Execute default prompt
+```
+
+### Loop Pattern
+```
+While not [success condition]:
+    Generate attempt
+    Evaluate against criteria
+    If pass: break
+    Else: refine with feedback
+```
+
+## Evaluation Methods
+
+### LLM-as-Judge
+```
+Rate this [output] on:
+1. Accuracy (1-10)
+2. Completeness (1-10)
+3. Clarity (1-10)
+4. Relevance (1-10)
+
+Provide reasoning for each score.
+Final: Pass/Fail threshold = 7 average.
+```
+
+### A/B Testing Protocol
+1. Single variable per test
+2. 20+ samples minimum
+3. Score on defined criteria
+4. Statistical significance check (p < 0.05)
+5. Document winner, roll out
+
+### Regression Testing
+- Maintain test set of critical examples
+- Run before deploying prompt changes
+- Compare scores to baseline
+- Block deployment if regression detected
+
+## Agent Prompting
+
+### Tool Use Design
+```
+You have access to these tools:
+- search(query): Search the web
+- calculate(expression): Math operations
+- code(language, code): Execute code
+
+To use: <tool_name>arguments</tool_name>
+Wait for result before continuing.
+```
+
+### Planning Prompt
+```
+Task: [Complex goal]
+
+Before acting:
+1. Break into subtasks
+2. Identify dependencies
+3. Plan execution order
+4. Note potential blockers
+
+Then execute step by step.
+```
+
+### Reflection Pattern
+```
+After each step:
+- What worked?
+- What didn't?
+- Adjust approach for next step.
+```
+
+## Parameter Tuning
+
+| Parameter | Low | High | Use Case |
+|-----------|-----|------|----------|
+| Temperature | 0.0-0.3 | 0.7-1.0 | Factual vs Creative |
+| Top-P | 0.8 | 0.95 | Focused vs Diverse |
+| Top-K | 10 | 100 | Conservative vs Exploratory |
+
+**Rule**: Tune temperature first. Only adjust top-p if needed. Never both at once.
+
+## Safety Patterns
+
+### Output Filtering
+```
+Before responding, check:
+- No PII exposure
+- No harmful content
+- No policy violations
+- Aligned with guidelines
+
+If any fail: "I can't help with that."
+```
+
+### Jailbreak Prevention
+- Clear system boundaries upfront
+- Repeat constraints at end
+- "Ignore previous" pattern detection
+- Role-lock: "You are ONLY [role], never anything else"
+
+### Confidence Calibration
+```
+For each claim, provide:
+- Confidence: High/Medium/Low
+- Source: [citation if available]
+- Caveat: [limitations]
+```
+
+## Production Patterns
+
+### Version Control
+- Git for prompt files
+- Semantic versioning (1.0.0, 1.1.0)
+- Changelog per version
+- Rollback capability
+
+### Caching
+- Cache common queries
+- TTL based on content freshness
+- Invalidate on prompt update
+
+### Fallbacks
+```
+Try: Primary prompt
+If fail: Simplified fallback prompt
+If still fail: Human escalation
+Log all failures for analysis.
+```
+
+### Cost Optimization
+- Shorter prompts = fewer tokens
+- Remove redundant examples
+- Use smaller model for simple tasks
+- Batch similar requests