create-skill

---
name: create-skill
description: >
  Use when Christopher wants to design, draft, or build a new Claude Code skill — either for
  this repo or for makemyskill.com. TRIGGER when: "create skill", "build skill for", "make a skill",
  "skill template", "new skill", "skill for [task]", "draft a skill". Do NOT use for: general
  task execution, explaining how skills work conceptually, or when Christopher asks what skills
  exist (read the .claude/skills/ directory for that).
---

# Skill: /create-skill

Design a new Claude Code skill file. Produce a draft ready for review and optionally for
submission to makemyskill.com ("describe the skill — the longer the better").

## The 7 Rules for Effective Skills

1. **Debug with description echo:** Ask Claude "When would you use the [skill-name] skill?" — it quotes the description back. Instantly reveals what's vague or missing.
2. **Negative triggers matter more than positive ones:** The "Do NOT use for..." line prevents skill hijacking. Write it first.
3. **Skills stack with CLAUDE.md and memory:** The skill handles *process*. CLAUDE.md handles *context and rules*. Memory handles *voice and history*. Don't repeat rules from CLAUDE.md inside the skill.
4. **Build from past conversations:** Don't start from scratch. The best skill prompts are reverse-engineered from prompts that already worked well in past sessions.
5. **Description is everything:** If too vague, the skill never fires. If too broad, it hijacks conversations it shouldn't. Test both failure modes before finalizing.
6. **Laziness workaround:** If Claude cuts corners inside a skill, don't edit the skill — add "Take your time. Quality over speed. Don't skip steps." to the invocation prompt.
7. **Skills are portable:** The SKILL.md format is designed as an open standard. Build for Claude today; designed to transfer across platforms.

## Skill File Format

```markdown
---
name: skill-name
description: >
  One or two sentences. TRIGGER when: [specific trigger conditions].
  Do NOT use for: [anti-triggers — be specific].
---

# Skill: /skill-name

[Full instructions. Be thorough — the longer the better for makemyskill.com.
Only the 3-line header is read at context load; full body only loads when triggered.]

## Before Starting
[What to check/read before executing]

## Steps
[Numbered or sectioned process]

## Output Format
[What the output should look like]

## After Completing
[Cleanup, commits, follow-on actions]
```

## Generation Workflow

### 1. Write the Description First
**Start with the "Do NOT use for" line** — prevents skill hijacking before positive triggers are written.

Then write the positive triggers:
- What does it do? (One sentence)
- TRIGGER when: [list specific keywords, task types, file types, user phrasings]
- Include synonyms and adjacent terms — Claude tends to undertrigger, so be "pushy"

### 2. Write the Body — Be Ruthlessly Concise
Every line must earn its place.
- **Explain WHY, not just WHAT** — Claude responds better to reasoning than rigid rules
- **Only teach what Claude doesn't know** — don't restate obvious best practices
- **Concrete examples beat abstract placeholders**
- Target: 150–350 lines total across all files

### 3. Self-Review Before Saving
- Can I cut 20% without losing value? Do it.
- Are there rigid rules that should be reasoning instead?
- Does the description cover all trigger scenarios?
- Would this work across many different prompts?

### 4. Test the Trigger
Ask: "When would you use the [skill-name] skill?" — Claude quotes the description back.
Reveals what's vague, too broad, or too narrow.

### 5. Run Through makemyskill.com (Before Shipping)
Submit just the `description` section — the longer the better. Replace description with refined output. Re-test.

### 6. Save and Register
- Save to `.claude/skills/skill-name.md`
- Add to MEMORY.md skills table
- Add to `setup/system-overview.md` Skills Library + Roadmap table
- Update CLAUDE.md skills table (trading-assistant and intent repos only)

## How to Use This Skill

1. Christopher describes what he wants the skill to do
2. Fortuna drafts the full skill file following the format above
3. Christopher reviews — test the description with the debug trick (ask "When would you use X?")
4. Optionally: submit the description section to makemyskill.com for a refined version
5. Save the final file to `.claude/skills/skill-name.md`

## Quality Checklist

- [ ] Description covers WHAT + WHEN, includes triggers and anti-triggers
- [ ] "Do NOT use for" is more specific than positive triggers
- [ ] Body is under 300 lines, focused and actionable
- [ ] Explains WHY behind guidance, not just rigid rules
- [ ] Every line earns its place — no generic filler
- [ ] No sensitive data, secrets, or account numbers (see Version Control Safety)
- [ ] `## Quick Commands` section present with exact bash commands
- [ ] Trigger tested with description echo before committing

## Strong vs. Weak Descriptions

**Weak:**
```yaml
description: A skill for building dashboards.
```

**Strong:**
```yaml
description: >
  Build simple, fast dashboards to display data. Use this whenever the user
  mentions dashboards, data visualization, metrics display, charts, graphs,
  or wants to display any kind of data visually, even if they don't explicitly
  ask for a 'dashboard'. Do NOT use for: complex BI tools, data warehousing,
  or ETL pipelines.
```

The strong version answers: **What** (build dashboards), **When** (data viz, metrics, charts), **What Not** (BI, warehousing, ETL).

## Skills Roadmap (Next to Build)

Candidates in priority order — run each through makemyskill.com before writing the full body.
Full roadmap with descriptions: `setup/system-overview.md` → Skills Candidates Roadmap section.

| # | Skill | What It Does |
|---|-------|-------------|
| 1 | `/level-brief` | Read TradingView MCP pine_labels/lines → summarize key levels per instrument for the session ✅ Built |
| 2 | `/smt-scan` | Cycle pane_focus NQ/ES/YM → identify index divergences → Scenario A/B/C verdict ✅ Built |
| 3 | `/session-sync` | Commit + push + append AGENT_SYNC.md + create/update session log ✅ Built |
| 4 | `/marp-deck` | Convert any review/briefing doc to Marp slide deck + generate HTML ✅ Built |
| 5 | `/import-trades` | Run TradeZella → STB pipeline, verify output, flag missing reviews ✅ Built |
| 6 | `/prop-firm-status` | Live account data → balance, floor, gap to target, daily-run-rate math ✅ Built |
| 7 | `/open-orders` | Pull positions/orders from all connected exchanges → single status report ✅ Built |
| 8 | `/pattern-review` | Read pattern_tracker.md → analyze trends → append dated review block with fix per pattern ✅ Built |
| 9 | `/monthly-audit` | End-of-month Marp deck: P&L + behavioral arc + strategy/course progression for all coaching groups ✅ Built |
| 10 | `/capture-agent-trades` | Future skill — capture and log trade data when Fortuna executes trades autonomously on Christopher's behalf 🔜 Planned |
| 11 | `/tax-entry` | Append trade or document to `taxes/YYYY/` working files ✅ Built |
| 12 | `/smog-analysis` | SMOG framework analysis for a specific trade or setup — ref `strategies/inevitrade/smog-reference.md` ✅ Built |
| 13 | `/tcl-analysis` | TCL strategy analysis for a trade or setup — ref `strategies/inevitrade/tcl-reference.md` ✅ Built |

## Version Control Safety

Skills in `.claude/skills/` are tracked in git and can appear in version control history.
**Before committing any skill file, verify it contains none of the following:**

| Risk | Examples | What to do instead |
|------|----------|--------------------|
| **Plaintext secrets** | API keys, tokens, passwords, seed phrases | Use `[YOUR_API_KEY]` placeholder |
| **Account numbers** | Prop firm IDs, exchange account IDs | Use `[ACCOUNT_ID]` or generic label |
| **Personal file paths** with sensitive info | Hardcoded `/Users/[name]/private/...` paths | Use relative paths or `~/` prefix |
| **Credential patterns** | `Authorization: Bearer abc123...` | Show shape only, never real values |
| **Sensitive history** | Trade details, P&L figures, coaching notes | These belong in `smarttrader-ai/` not skills |
| **Automatic ingestion** | Skill that reads `.env`, keychain, or private dirs on trigger | Only read files Christopher explicitly points to |

**Quick Commands sections** follow the same rules: use `[PLACEHOLDER]` for anything that
changes per account or person. Actual values (API keys, ngrok URLs, account numbers) live
in `.env`, Keychain, or private config — never in the skill file itself.

**If a skill needs sensitive context to work:** document the shape of what's needed
(e.g. "requires BTCC API key — store in Keychain as `btcc-api-key`") and let the
session provide it at runtime, not the skill file.

## Marp Slide Option

For any skill that produces a shareable document (coach notes, session briefings, weekly decks):
Use `/marp-deck` to generate the slide version. Public skill decks go in `setup/` — not in
`.claude/skills/`. See `setup/create-skill.marp.md` for the canonical template.
[▶️ Rendered view](https://drasticstatic.github.io/trading-assistant-public-preview/setup/create-skill.marp.html)

## Quick Commands

```bash
# Save a new skill file (gitignored — local only)
# File path: .claude/skills/skill-name.md

# After saving, test the description trigger:
# Ask Fortuna: "When would you use the [skill-name] skill?"

# Update MEMORY.md skills table and system-overview.md Skills Library
# Then run through makemyskill.com before shipping

# For skills that include Marp decks:
marp setup/create-skill.marp.md -o setup/create-skill.marp.html
git add setup/*.marp.md setup/*.marp.html && \
  git commit -m "Update [skill-name] Marp deck"
git push origin main
```