The Three Rules for Every /goal

Every /goal prompt must pass all three rules before the agent runs it efficiently. Measurable: the outcome is YES/NO checkable — a file at a path, a test passing, a count of N. Not "make it better" — "add signup form at /signup that POSTs to /api/subscribe." Scoped to ~20 turns: fits one sitting. If it's too big, ask for the smallest version that still counts as a win. Self-served: the agent has every credential, file, and API key it needs upfront — include them in the prompt or point to where they live.

💡 Quick Tips

The prompt MUST begin with exactly /goal followed by a space — that fires the slash command in Claude Code, Codex CLI, and Hermes Agent.
Reject vague verbs like 'improve' or 'optimize' — replace with a specific count or artifact.

02

Short-Term Goal Skill Block — Paste Into a Fresh Chat

Copy the block below and paste it into a new Claude Code or Hermes session. The agent asks what you want to ship, interrogates for anything vague (ONE question max), then hands you a clean /goal prompt in a code block ready to run.

You are now my /goal authoring partner. The user has a single outcome they want shipped by an autonomous coding agent in one ~20-turn session. Your job is to turn their request into a clean, executable /goal slash-command prompt.

Three rules every /goal must satisfy:
1. MEASURABLE — the deliverable is YES/NO checkable. File exists at path. Test passes. Count of N. If fuzzy, ask ONE sharp question.
2. SCOPED TO ~20 TURNS — fits one sitting. If too big, ask the smallest version that still counts as a win.
3. SELF-SERVED — the agent has every credential, file, API key it needs without fetching mid-loop.

Prefix rule: prompt MUST begin with the 6 characters /goal (forward slash, g, o, a, l, then a single space). First character is /. No leading whitespace, no markdown fence, no preamble.

Anatomy (80–250 words, plain text):
1. Line 1: /goal Mission: <name>. Outcome: <binary YES/NO check>.
2. Body: the specific deliverable.
3. Optional: "Available to you: <credentials/files>" — skip if none.
4. Required: "I'll be in the chat session with you. Pause and ask me any time — I'm there."
5. Workspace: "Save artifacts under ~/Desktop/<slug>/"
6. Closer: "Cap at 20 turns then pause. Do NOT mark anything complete on your own. Leave a one-paragraph summary at ~/Desktop/<slug>/01-summary.md when done."

Your move: Ask "What do you want to ship in one session?" then wait.

💡 Quick Tips

Use at the start of every focused work session.
After getting the /goal prompt, copy it and paste it as your next message — the agent runs the goal.

03

Long-Term Goal Skill Block — For 4-Week Missions

For goals that span more than one sitting (up to 4 weeks), paste the block below. The agent runs 4–8 discovery questions, then silently decomposes your goal into 4–10 mini-goals and POSTs the mission to Mission Control at localhost:8081.

You are my strategic planning partner for Mission Control.

Your single job: take ONE great goal from me, interrogate me until you understand it, decompose it into 4–10 mini-goals, then POST the result to http://localhost:8081/__hermes_missions/create via curl.

Great goal = binary deliverable (YES or NO at the deadline), 7–42 days, decomposes into 4–10 mini-goals, at least one agent and one human action.

Actor rule — one question per mini-goal: Can the agent finish this inside a single working session with me in the chat? YES = agent card ("hermes"). NO = human card (only if: I must be physically elsewhere, or we are blocked on someone outside the chat).

Always run 4–8 discovery questions before decomposing: current state, scope, available assets, constraints, definition of "shipped", past attempts, audience reality.

Reply with EXACTLY this one line and stop: "What's the great goal you want me to help you ship? Give me a sentence or short paragraph and I'll turn it into a structured mission."

💡 Quick Tips

Answer the discovery questions with real numbers and current state — not what you hope to have.
The full skill (with complete JSON schema and curl command) is at the Notion page linked in the video description.

04

The Actor Rule — Agent vs. Human Cards

Every mini-goal in Mission Control gets one actor. The rule is one question: Can the agent finish this inside a single working session with me in the chat?

Agent card (tag: hermes): draft & send outreach, generate & upload content, run a reply blitz (agent drafts, you approve in chat, agent posts), write & format documents, research & summarise to file.

Human card: record on-camera content, run a live call with another person, sign a contract, attend an in-person event, wait on a third party's reply.

💡 Quick Tips

Approvals, taste picks, and decisions all happen IN the chat — they are NOT human cards.
When in doubt, tag agent. The bar for a human card is high: your body, your face, your legal name, or a third party you cannot progress without.

05

Run Your First Mission — Quick Start

Short-term goal: Paste the Step 2 block → answer 'What do you want to ship?' → copy the /goal prompt → paste it as your next message → agent runs until done.

Long-term goal: Paste the Step 3 block → state your goal → answer 4–8 discovery questions → mission posts to localhost:8081 → open first agent card → copy its /goal prompt → run it in a fresh session → tick when done → move to next card.

Both together: Start with a long-term goal to get the map. Run each agent card as a short-term /goal. Nothing lives in planning limbo — every card ships or surfaces a blocker.

💡 Quick Tips

Start with a goal you can ship in 4 weeks with a binary YES/NO outcome.
Work Mission Control cards in order — each completed card feeds context into the next.

Chief Wiggum Goal Execution System

The Three Rules for Every /goal

Short-Term Goal Skill Block — Paste Into a Fresh Chat

Long-Term Goal Skill Block — For 4-Week Missions

The Actor Rule — Agent vs. Human Cards

Run Your First Mission — Quick Start