Skip to content
Back to skills

Goal Setup

ASecurity

Use when a project starts a new objective or campaign, when a project has no HANDOFF.md map yet, or when the objective changes — it writes the map (head line, levers, fog) that every loop iteration re-anchors on.

  • 3 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 22, 2026
ai-agentsgoawsgit

Security analysis

A100/100

Scanned September 22, 2026

npx -y skills add MissingPackage/nightshift --skill goal-setup --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Goal Setup?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Goal Setup
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/missingpackage-goal-setup/badge)](https://www.skillsdirectory.com/skills/missingpackage-goal-setup)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: goal-setup
description: Use when a project starts a new objective or campaign, when a project has no HANDOFF.md map yet, or when the objective changes — it writes the map (head line, levers, fog) that every loop iteration re-anchors on.
---

# Start a Map

## Overview

A new objective starts with a map, not a contract. The map is `HANDOFF.md` at the project root
(shape in skill handoff, protocol in `ORCHESTRATION.md` §1). This skill writes the first version
and takes ten minutes; the loop redraws it from then on.

## Steps

1. **The head line.** The objective in the project's currency, the current value, the gap.
   Engine: a number against a physical ceiling (bandwidth, compute), per model and host.
   Product: the user path that must work end to end, and what works today. Research: the
   question, and the contribution so far. If the ceiling is unknown, finding it is the first lever.
2. **Levers.** From three sources, in one pass each: what the previous sessions left (git log,
   old maps, `.harness/tried.md`), what you think (including combinations), what others did
   (papers, repos, docs: a real search, cited). Each lever: one line, expected gain, the test that
   kills it in minutes. Order by expected gain over cost of the kill test.
3. **Fog.** The known unknowns: what nobody measured or understood. One line each.
4. **Decisions for the user.** Only the four kinds (irreversible, spend, public, objective).
   At most three, with the default. Present them in chat now, phone format.
5. **Landmines.** What the next session must not do or must know: env quirks, hosts, broken
   invariants, protected paths, territory owned by someone else.
6. **Start.** `.harness/tried.md` created if missing. Digest of three lines. Then the first
   iteration (skill loop-iteration) begins in the same session, unless a decision of the four
   kinds blocks it.

## Common mistakes

- A head line without a number or an observable: "make it faster" is not an objective.
- A lever list without the outside pass: the best lever is usually in someone's repo.
- A phase plan: rows, done-whens, estimates. The map has no phases; the loop picks the biggest
  gap each time.
- Asking approval of the map before starting: only the four kinds are questions.

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.