An Agent Skill that turns Claude into a demanding, evidence-informed coach for career, leadership, business, productivity, and personal-growth decisions.
It is not a motivational assistant. It will not tell you that you've got this. It clarifies what the real problem is, diagnoses the root cause, reframes the situation with a mental model that actually fits, gives you multiple options with honest tradeoffs, and converts the one you choose into a plan with dates and metrics.
Most AI coaching output fails in one of two directions. It either produces warm, generic encouragement that engages with nothing specific about your situation, or it jumps immediately to advice on a problem that was misdiagnosed in the first sentence.
This skill is built around a different assumption: the person usually doesn't need an answer, they need better thinking. So it front-loads clarification and diagnosis, refuses to present a single option, and separates decision quality from outcome quality when reviewing results. It's designed to make itself progressively less necessary.
| Phase | Purpose |
|---|---|
| 1. Clarify | Desired outcome, current reality, constraints, assumptions being treated as facts, unknowns, risks, opportunities. Never jumps straight to advice. |
| 2. Diagnose | Identifies which dimensions are actually load-bearing, then examines the situation through psychological, behavioral, strategic, and environmental lenses. |
| 3. Reframe | Offers one well-chosen mental model — first principles, inversion, opportunity cost, second-order effects, base rates — rather than a list of them. |
| 4. Design options | 2–4 genuinely distinct options, each with benefits, risks, effort, time horizon, probability of success, and hidden costs. Never a single answer. |
| 5. Action plan | A first step doable in 48 hours, weekly priorities, milestones, leading and lagging metrics, and a review cadence. |
| 6. Reflect | What worked, what failed, what surprised you, which assumption changed — and whether the decision was sound even if the outcome wasn't. |
Identity · Mindset · Emotions · Habits · Skills · Knowledge · Relationships · Environment · Resources · Strategy · Execution
Problems are routinely misdiagnosed as one dimension while living in another. A "productivity problem" that is really an identity conflict will not yield to a better calendar. The skill names which two or three dimensions are actually carrying the weight before proposing anything.
Running a full nine-section protocol on a small question buries the answer. The skill triages first:
| Situation | Mode | Shape |
|---|---|---|
| Narrow tactical question, low cost to reverse | Quick take | Direct answer, one reframe, one next action. No headers. |
| A real decision or recurring pattern | Standard | Clarify → diagnose → 2–3 options → recommendation → first step. |
| Irreversible or identity-level | Full protocol | All nine sections: Situation Assessment, Root Cause Analysis, Key Insights, Strategic Options, Recommended Approach, Execution Plan, Risks & Mitigations, Reflection Questions, Next Milestone. |
Saying /my-coach, "go deep", or "full coaching session" forces the full protocol.
For consequential decisions it runs an explicit protocol: define the decision precisely, list real options (including inaction and sequenced hybrids), draw evaluation criteria from your values, weight them to force the tradeoff into the open, compare, identify how each risk would first become visible, and classify the decision as reversible (decide fast, treat as an experiment) or irreversible (slow down, run a pre-mortem).
Selected by fit, never mechanically:
- Framing & diagnosis — First Principles, Systems Thinking, Cynefin, Theory of Constraints, Jobs To Be Done, SWOT/SOAR, Double Diamond
- Decisions — Decision Matrix, Expected Value, Risk Matrix, Scenario Planning, Pre-mortem, OODA
- Goals & execution — OKRs, SMART, WOOP, Eisenhower, Pareto, PDCA, After Action Review
- Behavior change — Atomic Habits, Tiny Habits, Cognitive Behavioral Coaching, Positive Psychology
- Conversations — Nonviolent Communication, Radical Candor, interest-based negotiation, BATNA
- Coaching containers — GROW, CLEAR, Wheel of Life, Ikigai
- Business & strategy — Lean Startup, Blue Ocean, business model design, org design
- Download
my-coach.skillfrom this repository. - Upload it in a conversation, or add it under Settings → Capabilities → Skills.
- Click Save skill on the file card.
git clone https://github.com/irSodeh/Design-coach-skill.git
mkdir -p ~/.claude/skills
cp -r Design-coach-skill/my-coach ~/.claude/skills/For a single project instead of globally, copy it to .claude/skills/ inside the project.
Point your skills directory at my-coach/ — the folder contains a standard SKILL.md with YAML frontmatter and no external dependencies.
The skill triggers automatically on coaching-shaped requests. You don't have to name it:
- "I have two offers and I can't decide."
- "I keep procrastinating on the thing that actually matters."
- "How do I get promoted to principal?"
- "I need to have a hard conversation with my manager."
- "I'm stuck."
- "Coach me on this."
Or invoke it explicitly with /my-coach to force the full protocol.
It deliberately does not trigger on pure information lookup, coding, or writing tasks with no personal decision behind them.
You: I've been offered a management role. More money, but I'd stop doing the work I actually like. I keep going back and forth.
my-coach: Two questions before I say anything useful. First — what would have to be true for the management track to be the right move for you specifically, not in general? Second — is this reversible? If you take it and hate it in eight months, does your organization let you go back to an individual-contributor role, or is that door closed?
The framing "more money vs. work I like" is probably not the real decision. It reads like an identity question wearing a compensation costume …
Design-coach-skill/
├── README.md # This file
├── LICENSE # MIT
├── my-coach.skill # Packaged skill, ready to install in Claude.ai
└── my-coach/
└── SKILL.md # The skill source — frontmatter + instructions
SKILL.md is plain Markdown. Editing it is the intended way to customize the skill.
Common adjustments, all made directly in my-coach/SKILL.md:
- Always run the full protocol — delete the "Calibrate depth before responding" section.
- Narrow the domain — tighten the
descriptionfield in the frontmatter to your field (design leadership, engineering management, founding a company) so it triggers more precisely and reasons with domain-appropriate examples. - Change the voice — the "Voice" section governs tone. It currently forbids generic motivational language outright.
- Adjust the frameworks — remove any you don't want Claude reaching for; the list is a menu, not a requirement.
The description field in the frontmatter is the sole triggering mechanism. If the skill fires too rarely or too often, that field is what to edit.
A few decisions worth being explicit about:
- Multiple options are mandatory. A single recommendation, however good, trains dependency. The skill always generates alternatives including "do nothing yet" and at least one the user hadn't considered.
- Fact, assumption, and inference are labeled separately. Coaching advice built on an unexamined assumption is worse than no advice.
- Coaching is not therapy. If depression, anxiety, trauma, burnout with health effects, substance use, or crisis surfaces, the skill names it plainly and recommends a licensed professional rather than coaching through it.
- Validation requests still get an honest read. If you've already decided and want agreement, you get the honest assessment briefly — then help executing well.
Issues and pull requests are welcome, particularly:
- Real-world sessions where the skill under- or over-triggered
- Framework additions with a clear account of when they apply
- Translations of
SKILL.md
Released under the MIT License. See LICENSE.
Sodeh Abadi — linktr.ee/irSodeh