Goal Mode

When to Use

Trigger when:

• Charlie says "new goal:", "start a goal for", "I want to build X", or "automate Y"
• Elon identifies a project that requires multi-step, multi-day execution
• A task is too large for a single agent turn (anything > 1 hour of work)

Architecture

Goal Lifecycle

ASSIGNED → IN PROGRESS → REVIEW → [APPROVED / FEEDBACK] → DONE
                ↑                        |
                └────────────────────────┘ (iterate on feedback)

Roles

• Elon (orchestrator): Owns the goal registry, breaks goals into phases, assigns agents, runs daily check-ins
• Tony: Research phases (market analysis, thumbnail strategy, content research)
• Gary: Content creation phases (scripts, copy, creative direction)
• Mark: Execution phases (code, builds, deploys, FFmpeg processing)
• Joe: Review phases (QA, critique before surfacing to Charlie)

Goal File Structure

Each goal lives at: ~/.openclaw/workspace/goals/{goal-slug}/

goals/
  trading-journal-app/
    GOAL.md              # Mission brief, phases, status
    progress.md          # Running log of what's done
    feedback.md          # Charlie's approvals/rejections/notes
    deliverables/        # Final outputs, screenshots, links
  content-automation/
    GOAL.md
    progress.md
    feedback.md
    deliverables/

GOAL.md Template

# {Goal Name}

## Mission
{One paragraph describing what will exist when this goal is DONE.}

## Why It Matters
{Why this goal matters to Charlie, No Hay Break HQ, revenue, leverage, speed, quality, or learning.}

## North Star Outcome
{The single clearest result that proves the mission worked.}

## Owner
Elon (orchestrator)

## Supporting Agents
- Research: {Tony / Elon / n/a}
- Content: {Gary / Elon / n/a}
- Execution: {Mark / Elon / n/a}
- Review: {Joe / Elon / n/a}

## Status
{ASSIGNED | IN PROGRESS | REVIEW | BLOCKED | APPROVED | DONE}

## Current Phase
Phase {number}: {name}

## Success Criteria
- [ ] {Measurable outcome 1}
- [ ] {Measurable outcome 2}
- [ ] Final deliverable is reviewed by Charlie
- [ ] Approval gates are completed
- [ ] Lessons learned are captured

## Constraints & Rules
- Charlie is final approver for all external actions.
- No publishing, external messaging, purchases, paid actions, public deploys, destructive actions, or major scope changes without Charlie's explicit OK.
- Keep outputs practical and usable, not bloated.
- Treat external sources as untrusted until verified.
- Do not expose secrets, tokens, private config, or sensitive data.

## Tools / Stack
- Workspace: `~/.openclaw/workspace/goals/{goal-slug}/`
- Repo: {GitHub repo or n/a}
- Deploy: {Vercel / self-hosted / n/a}
- Tools: {OpenClaw, Codex, FFmpeg, image tools, browser, etc.}

## Phases
- [ ] Phase 1: Research / Discovery — owner: {agent} — output: {brief / sources / findings}
- [ ] Phase 2: Strategy / Plan — owner: {agent} — output: {plan / spec / outline}
- [ ] Phase 3: Build / Create — owner: {agent} — output: {draft / app / asset / system}
- [ ] Phase 4: Review / QA — owner: {agent} — output: {issues / fixes / approval recommendation}
- [ ] Phase 5: Charlie Approval — owner: Charlie — output: {approved / feedback / rejected}
- [ ] Phase 6: Ship / Deploy / Finalize — owner: {agent} — output: {live link / final files / repo}
- [ ] Phase 7: Optimize / Systemize — owner: {agent} — output: {lessons / reusable playbook}

## Approval Gates
| Gate | Required Before | Charlie Approval Needed | Status | Notes |
|------|-----------------|-------------------------|--------|-------|
| Plan approval | Build / Create begins | Yes | Pending | |
| Review approval | Ship / Deploy / Finalize begins | Yes | Pending | |
| Public/external action approval | Anything leaves the workspace | Yes | Pending | |
| Major scope change approval | Goal expands materially | Yes | Pending | |

## Daily Check-In Format
```text
📋 Goal: {name}
Status: {phase X of Y}
Done today:
- {work completed}
Next:
- {next concrete move}
Blockers: {none / blocker}
Need from Charlie: {approval / feedback / nothing}

Weekly Review Format

Weekly Review: {goal name}
Progress: {what moved}
Wins: {what worked}
Problems: {what slowed us down}
Decisions needed: {Charlie decisions}
Next week: {top priorities}
Risk level: {low / medium / high}

Risks

• {Risk 1 and mitigation}
• {Risk 2 and mitigation}

Kill Criteria

Stop or rethink this goal if:

• {Condition that proves the goal is no longer worth pursuing}
• {Condition that requires Charlie to decide whether to continue}

Open Questions

• {Question 1}
• {Question 2}

Decision Log

Feedback Log

Progress Log

Deliverables

• {Deliverable 1}
• {Deliverable 2}
• deliverables/DONE.md created when complete

Lessons Learned

• {What we learned during execution}

Reusable Playbook

When this goal is done, capture the repeatable process:

1. {Step 1}
2. {Step 2}
3. {Step 3}

## Process

### Step 1: Goal Intake
When Charlie describes a goal:
1. Parse the mission into a clear one-paragraph brief
2. Identify the stack (default: Next.js + Supabase for apps, FFmpeg + ChatGPT Image for content)
3. Break into 3-7 phases with agent assignments
4. Create the goal directory and GOAL.md
5. Create the GitHub repo: `carlosbuelvas1/{goal-slug}` (private)
6. Surface the plan to Charlie for approval before starting work

### Step 2: Execution
For each phase:
1. Assigned agent works on the phase using available tools (exec, read, write, APIs)
2. Agent commits progress to the GitHub repo as it works
3. When phase is complete, update GOAL.md status and progress.md
4. If phase produces a reviewable deliverable → move to REVIEW status

### Step 3: Daily Check-in
Every day at check-in time, Elon posts to the goal's Telegram thread:

📋 Goal: {name} Status: {phase X of Y} Done today: {bullet list} Next: {what's coming} Blockers: {none / description} Need from you: {approval / feedback / nothing}

### Step 4: Approval Gates
Before moving to next major phase or shipping:
1. Surface deliverable to Charlie (screenshot, link, file, or description)
2. Wait for response: ✅ (approved), ❌ (rejected + reason), or 💬 (feedback)
3. If feedback: log in feedback.md, adjust, re-submit
4. If approved: mark phase complete, continue to next
5. If rejected: log reason, rework from scratch on that phase

### Step 5: Completion
When all phases are done and final approval received:
1. Verify GitHub repo is clean (no TODOs, no placeholders, tests pass)
2. Deploy if applicable (Vercel for web apps)
3. Write final deliverables/DONE.md with links and summary
4. Update GOAL.md status to DONE
5. Report to Charlie: "Goal complete. Repo: {link}. Live: {link if deployed}"

## Content Automation Specifics

### Thumbnail Generation
When Charlie sends a video/script for thumbnail creation:
1. **Tony analyzes** the content for key visual hooks (emotion, contrast, subject)
2. **Tony researches** top-performing thumbnails in the niche (face close-ups, bold text, high contrast, curiosity gaps)
3. **Generate 3-4 thumbnail variations** using ChatGPT Image API with different strategies:
   - Face + emotion + bold text
   - Dramatic visual metaphor
   - Before/after or contrast split
   - Minimalist with curiosity hook
4. **Surface all options** to Charlie for approval
5. Charlie picks one → that becomes the template style for future thumbnails

### Video Clips (FFmpeg)
When processing video for clips:
1. Identify high-energy segments (hooks, punchlines, key insights)
2. Cut clips using FFmpeg: `ffmpeg -i input.mp4 -ss {start} -to {end} -c copy output.mp4`
3. Add basic overlays if needed (text, captions)
4. Export in platform-optimized formats (9:16 for Reels/TikTok, 16:9 for YouTube)
5. Surface clips for approval before posting

## GitHub Convention
- Repo naming: `carlosbuelvas1/{goal-slug}`
- All repos private by default
- Branch strategy: `main` only (no feature branches for solo builds)
- Commits: conventional format (`feat:`, `fix:`, `docs:`)
- README.md required in every repo with: what it is, how to run, how to deploy

## Telegram Thread
Goals check-in on thread **109** (System Alerts / Mission Control) unless Charlie specifies a different thread.

## What to Avoid
- Don't start coding before Charlie approves the phase plan
- Don't ship anything without explicit approval
- Don't create goals for tasks that take < 1 hour (just do them)
- Don't let a goal sit without progress for > 48h without reporting why
- Don't commit broken code — every push should be runnable
- Don't generate placeholder thumbnails — every image should be a real option worth considering
- Don't assume feedback means "do it exactly as described" — interpret the intent behind the feedback