SKILL_MODE Pattern — v1.0

SKILL_MODE Pattern — v1.0

**Stolen from:** everything-claude-code (ECC) runtime profile system **Applied to:** AI marketing skills

---

What This Is

Every skill now supports three execution depths. Users get the right output for their situation without the agent guessing.

Mode	Output depth	Time expectation
quick	Minimum viable output — single pass, no deep research, no scoring	< 15 min
standard	Full process — all phases, scoring where relevant, priority output	30–45 min
deep	Extended research + iteration + frameworks for ongoing use	60–90 min

---

How to Add This to Any Skill

**Step 1: Add the Mode section** immediately after the intro, before "Before You Begin" or Phase 1.

## Mode

Detect from context or ask: *"[One-line question specific to this skill]"*

| Mode | What you get | Best for |
|------|-------------|----------|
| `quick` | [Describe the minimum useful output] | [When to use] |
| `standard` | [Full current skill behavior] | [When to use] |
| `deep` | [Extended output with extras] | [When to use] |

**Default: `standard`** — [one line on when quick or deep is appropriate]

**Step 2: Tag phases/sections** that only run in certain modes.

• No tag = runs in all modes
• [standard+] = runs in standard and deep only
• [deep only] = deep mode only

**Step 3: Update Output Format section** to describe what each mode delivers.

---

Skills Updated (v2.1)

• ai-discoverability-audit — quick (Phase 1 only) / standard (full) / deep (+ 90-day plan)
• cold-outreach-sequence — quick (1 message) / standard (4-touch) / deep (campaign system)
• de-ai-ify — quick (fast pass) / standard (full scan) / deep (+ voice calibration)

Skills To Update (Next)

• linkedin-authority-builder
• linkedin-profile-optimizer
• content-idea-generator
• homepage-audit
• reddit-insights
• last30days
• positioning-basics
• marketing-principles

---

Design Principles

1. quick is not worse — it's appropriate for different jobs. Don't frame quick as a degraded mode.
2. Don't change the skill's core logic. The mode gate is additive — standard mode = current behavior.
3. Detect from context first. If the user says "quick" or "fast" or gives you one name — infer quick. If they say "full audit" or "system" — infer deep. Only ask if genuinely ambiguous.
4. deep should earn its time. If deep mode doesn't produce something a consultant would charge for, it's not deep enough.

---

*Pattern introduced: March 8, 2026* *Inspired by: ECC's `ECC_HOOK_PROFILE=minimal|standard|strict`*