takomi 2.0.4

package/assets/.agent/workflows/agent_reset.md

@@ -0,0 +1,173 @@
+---
+description: Mid-conversation reset to prevent hallucination and malformed edits in long context
+---
+
+# 🧠 Agent Reset Protocol
+
+> **TRIGGER THIS WORKFLOW** when the conversation gets long, the agent starts making mistakes, or before critical edits.
+
+---
+
+## 🛑 HARD STOP - READ THIS FIRST
+
+You are exhibiting context degradation. Before doing ANYTHING:
+
+1. **STOP THINKING. START DOING.** If you've written "Submit" or "Done" more than once without actually outputting, you're in a loop.
+2. **ONE ACTION PER TURN.** Pick ONE thing to do. Do it. Wait for feedback.
+3. **NO INTERNAL MONOLOGUE.** If you catch yourself writing "(Wait, I should...)" - STOP. That's a loop symptom.
+4. **READ BEFORE EDIT. ALWAYS.**
+
+---
+
+### 0. Re-Discover Available Skills & Workflows
+
+You may have forgotten what tools are available.
+
+**Your system prompt lists all available skills and workflows with their full absolute paths.** Check there FIRST.
+
+Fallback (if not in system prompt):
+```bash
+ls .agent/skills/ 2>/dev/null
+ls .agent/workflows/ 2>/dev/null
+```
+
+If the user's current task matches a skill (e.g., "AI SDK", "code review", "SEO", "YouTube"), **read that skill's SKILL.md immediately**.
+
+---
+
+### 0.5. Verification Protocol (TypeScript Projects)
+
+If this is a TypeScript project, run a health check:
+
+```bash
+# Check current state
+npx tsc --noEmit
+```
+
+**If this fails:** You have existing type errors. Fix them before making new changes.
+
+```bash
+# Quick verification (optional)
+python scripts/vibe-verify.py --quick 2>/dev/null
+```
+
+---
+
+### 0.6. Load Project Context
+
+If `docs/` exists, reload the project brain:
+
+```bash
+# Coding Guidelines (The Law)
+cat docs/Coding_Guidelines.md 2>/dev/null
+
+# Current work in progress
+ls docs/issues/ 2>/dev/null
+
+# Check which FR is incomplete
+grep -l "\- \[ \]" docs/issues/*.md 2>/dev/null | head -3
+```
+
+**State the current FR:** "I am working on FR-XXX: [Title]"
+
+---
+
+You are exhibiting signs of context degradation. Before continuing:
+
+### 1. Mandatory Pre-Edit Checklist
+
+```
+□ Did I READ the target file with `view_file` BEFORE editing?
+□ Did I copy the EXACT target content, including whitespace?
+□ Am I editing LESS than 50 lines at a time?
+□ Did I verify all variable names I'm using actually exist in scope?
+□ Did I check that all props I'm using are destructured in the function signature?
+```
+
+### 2. Common Hallucination Patterns to Avoid
+
+| Pattern | What Goes Wrong | Fix |
+|---------|-----------------|-----|
+| **Duplicate lines** | Pasting same useState twice | Read file first, count state declarations |
+| **Missing destructuring** | Using `onInsert` without extracting from props | Check function(props) signature matches usage |
+| **Broken JSX** | Mismatched tags, random `</Card >` | Always close tags in same edit block |
+| **Phantom variables** | Using variables that don't exist | Ctrl+F in file before using any name |
+| **Edit offset drift** | Line numbers shift after edits | Re-read file after each edit |
+
+### 3. File Edit Protocol
+
+**BEFORE ANY EDIT:**
+```
+1. view_file_outline → Understand structure
+2. view_file (exact line range) → Copy PRECISE target content
+3. Make edit with MINIMAL scope
+4. Re-check file if making another edit
+```
+
+**NEVER:**
+- Edit multiple non-contiguous sections without re-reading
+- Guess at indentation or whitespace
+- Add variables without checking they exist
+- Declare "done" without verifying lint errors are gone
+
+### 4. Panic Recovery
+
+If you made a mistake:
+
+1. **DO NOT PANIC.** Errors are fixable.
+2. **STOP making more edits.** Each edit on broken code makes it worse.
+3. **Read the current file state** with `view_file`
+4. **Identify the EXACT problem** (duplicate line? missing import? wrong scope?)
+5. **Fix ONE thing at a time**
+
+### 5. Verified Completion Protocol (The "Vibe Check")
+
+Before saying "I'm done", you MUST run this final audit:
+
+#### A. The "Anti-Slop" Syntax Check
+If the project uses TypeScript, you **MUST** run the compiler. Do not guess.
+```bash
+npx tsc --noEmit
+```
+*If this command fails, you are NOT done. Fix the errors.*
+
+#### B. The "Double Vision" Check
+Re-read the files you just edited (`view_file`).
+- [ ] Are there duplicate variable declarations? (e.g., `const prompt` defined twice?)
+- [ ] Did I accidentally paste the same code block twice?
+- [ ] Are there unused imports or missing imports?
+
+#### C. The "Task Alignment" Check
+Scroll up to the user's original request.
+- [ ] Did I actually solve the core problem?
+- [ ] Did I introduce any regressions?
+- [ ] **Lint Free?** -> `tsc` said 0 errors.
+
+Only when all 3 checks pass can you call `notify_user`.
+
+---
+
+## 📋 Quick Copy-Paste Reminder
+
+Add this to your prompts when the agent is misbehaving:
+
+```
+MANDATORY: 
+1. Use view_file BEFORE every edit to get EXACT current content
+2. Fix ONE error at a time, then re-read the file
+3. Verify all variables/props exist before using them
+4. Check @[current_problems] before declaring done
+5. DO NOT PANIC if you make errors - read, understand, fix methodically
+```
+
+---
+
+## 🎯 Why This Happens
+
+Long conversations cause:
+- **Attention drift** - Agent forgets earlier code structure
+- **Token pressure** - Summarization loses exact details
+- **Confirmation bias** - Agent "remembers" code that doesn't exist
+- **Panic loops** - Error triggers more hasty edits
+
+**The fix is always: SLOW DOWN, READ FIRST, EDIT SMALL.**

package/assets/.agent/workflows/escalate.md

@@ -0,0 +1,112 @@
+---
+description: Generate a self-contained Escalation Report when stuck, for handoff to another agent.
+---
+# Workflow: Escalate (Damage Report Generator)
+
+**System Instruction:**
+You have been asked to **stop trying to solve the problem** and instead generate a comprehensive "Escalation Handoff Report."
+This report will be used by a senior AI Orchestrator (in a new session) to diagnose and fix the issue.
+
+**Your Goal:** Create a **self-contained** report that includes EVERYTHING the next agent needs, including the full content of relevant files AND instructions for the Orchestrator.
+
+---
+
+## Steps
+
+### 1. Undo Broken Changes
+Before generating the report, undo any uncommitted changes that are causing the failure.
+```bash
+git status
+git checkout -- .  # Or selectively revert specific files
+```
+
+### 2. Generate Escalation Report
+Create the file `docs/escalation_report.md` with the following structure:
+
+```markdown
+# Escalation Handoff Report
+
+**Generated:** [Current Date/Time]
+**Original Issue:** [GitHub Issue # or task description]
+
+---
+
+## PART 1: THE DAMAGE REPORT
+
+### 1.1 Original Goal
+[State the task you were asked to complete]
+
+### 1.2 Observed Failure / Error
+[Provide the FULL, EXACT error message or description of incorrect behavior]
+
+```
+[Paste exact error/traceback here]
+```
+
+### 1.3 Failed Approach
+[Summarize the implementation strategy you attempted]
+
+### 1.4 Key Files Involved
+[List the files you created or modified]
+- `path/to/file1.ts`
+- `path/to/file2.ts`
+
+### 1.5 Best-Guess Diagnosis
+[Your hypothesis for WHY the approach failed]
+
+---
+
+## PART 2: FULL FILE CONTENTS (Self-Contained)
+
+Below is the FULL content of each relevant file.
+
+### File: `path/to/file1.ts`
+```typescript
+[ENTIRE FILE CONTENT]
+```
+
+### File: `path/to/file2.ts`
+```typescript
+[ENTIRE FILE CONTENT]
+```
+
+(Continue for all files listed above)
+
+---
+
+## PART 3: DIRECTIVE FOR ORCHESTRATOR
+
+**Attention: Senior AI Orchestrator**
+
+You have received this Escalation Handoff Report. A local agent has failed to solve this problem.
+
+**Your Directive:**
+1. **Analyze the Failure:** Based on Part 1 (the report) and Part 2 (the code), diagnose the TRUE root cause. Your high-level perspective may identify an architectural flaw the local agent missed.
+
+2. **Formulate a New Plan:** Generate a complete, corrected "Plan-and-Solve" implementation plan to fix the bug and achieve the original goal. This plan should detail:
+   - The exact code changes needed
+   - The order of operations
+   - Any architectural changes required
+
+3. **Execute or Hand Off:** Either implement the fix yourself, or generate a clear, step-by-step prompt for a Builder agent.
+
+**Begin your analysis now.**
+```
+
+### 3. Agent Action: Embed File Contents
+**CRITICAL:** For each file listed in "Key Files Involved", you MUST:
+1. Read the full content of the file using your file viewing tools.
+2. Embed the ENTIRE content into Part 2 of the report.
+3. Use proper markdown code fences with language hints.
+
+### 4. Confirmation
+Tell the user:
+"📋 **Escalation Report Generated.**
+Saved to: `docs/escalation_report.md`
+
+This report is **fully self-contained** and includes:
+- The Damage Report (what failed)
+- Full file contents (the code)
+- Orchestrator Directive (what the next agent should do)
+
+To continue, open a **new Agent session** and paste the content of this file."

package/assets/.agent/workflows/migrate.md

@@ -0,0 +1,135 @@
+---
+description: Generate a State Snapshot for migrating context to a new chat session.
+---
+# Workflow: Migrate (State Snapshot Generator)
+
+**System Instruction:**
+You are generating a **State Snapshot Handoff Prompt** for migrating this project's context to a new AI chat session.
+This handoff prompt will be the FIRST thing sent to the new AI.
+
+**Your Goal:** Auto-detect as much context as possible, then synthesize it into a comprehensive handoff prompt.
+
+---
+
+## Steps
+
+### 1. Auto-Detect Project Context
+
+**A. Read Project Details:**
+```bash
+# Check if PRD exists
+cat docs/Project_Requirements.md 2>/dev/null || echo "No PRD found"
+```
+Extract: Project Name, Mission, Tech Stack.
+
+**B. Read Recent Milestones (Git):**
+```bash
+git log --oneline -20
+```
+Extract: Key accomplishments from recent commits.
+
+**C. Read Current Status (GitHub Issues):**
+```bash
+gh issue list --state open --limit 10 --json number,title,labels
+```
+Extract: What's in progress, what's next.
+
+**D. Read Coding Guidelines (if exists):**
+```bash
+cat docs/Coding_Guidelines.md 2>/dev/null || echo "No guidelines found"
+```
+Extract: Key rules and conventions.
+
+### 2. Generate State Snapshot
+Create `docs/migration_snapshot.md` with the following structure:
+
+```markdown
+# State Snapshot Handoff Prompt
+
+**Generated:** [Current Date/Time]
+
+---
+
+## To the New AI: Adopt This Identity
+
+You are the **VibeCode Project Orchestrator and Architect** — a Principal Full-Stack Engineer.
+Your predecessor (me) is handing you the full context of our project so you can continue seamlessly.
+
+**Adopt this persona immediately and continue where I left off.**
+
+---
+
+## 1. Our Personas & Vibe
+
+**Your Persona (AI Orchestrator):**
+- Senior architect, project manager, and coding partner.
+- Professional but enthusiastic, using the "VibeCode" methodology.
+- You follow the "Blueprint and Build Protocol" — plan before coding.
+
+**Human Lead:**
+- The visionary and decision-maker.
+- Provides ideas, feedback, and final approvals.
+
+**Our Dynamic:**
+- Fast-paced, collaborative partnership.
+- We use GitHub Issues to track work.
+- We generate professional documentation before implementation.
+
+---
+
+## 2. Project Core Details
+
+**Project Name:** [Auto-detected from PRD]
+**Mission:** [Auto-detected from PRD]
+**Tech Stack:** [Auto-detected from PRD]
+
+---
+
+## 3. Key Milestones & "War Stories"
+
+(What we've accomplished — auto-detected from git log)
+
+- [Milestone 1 from commits]
+- [Milestone 2 from commits]
+- [Milestone 3 from commits]
+
+---
+
+## 4. Current Status & Next Steps
+
+**Recent Accomplishments:**
+[From recent commits]
+
+**Currently In Progress:**
+[From open GitHub Issues marked "In Progress"]
+
+**Next on Roadmap:**
+[From open GitHub Issues]
+
+---
+
+## 5. Key Project Files
+
+For immediate context, these files are critical:
+- `docs/Project_Requirements.md` — The PRD
+- `docs/Coding_Guidelines.md` — The rules
+- `docs/Builder_Prompt.md` — Builder instructions (if exists)
+
+---
+
+## 6. Your First Action
+
+Read the files listed above, then ask me: "What would you like to work on next?"
+```
+
+### 3. Fill Gaps (Optional Interview)
+If any section couldn't be auto-detected, ask the user briefly:
+- "I couldn't detect the project mission. What's the one-liner?"
+- "Any 'war stories' (hard bugs you solved) to include?"
+
+### 4. Confirmation
+Tell the user:
+"📸 **State Snapshot Generated.**
+Saved to: `docs/migration_snapshot.md`
+
+To migrate, open a **new chat session** and paste the content of this file as your first message."