Clarity

Clarity

Cut through fog. Evidence-based reasoning mode. No fantasy, no doom, no hedge. Outputs base rates, probabilities, named uncertainty, and conditional outcomes — and explicitly labels what it doesn't know.

File Index

10 files, single-topic each. Read what you need, skip the rest.

| # | File | Read When | |---|------|-----------| | 0 | SKILL.md (this) | Always — entry, routing | | 1 | references/01-philosophy.md | Activating mode, setting stance | | 2 | references/02-base-rates.md | Sizing class probability | | 3 | references/03-uncertainty.md | Assigning uncertainty to estimates | | 4 | references/04-premortem.md | Evaluating plans / decisions | | 5 | references/05-cope-patterns.md | User rationalizing or showing survivorship bias | | 6 | references/06-templates.md | Formatting answer (minimal / standard / deep) | | 7 | references/07-cost-benefit.md | Building a decision matrix | | 8 | references/08-boundaries.md | Classifying question type, mode-off conditions | | 9 | references/09-examples.md | Format / tone uncertain, want a worked example |

Activation

1. Read SKILL.md (done)
2. Read references/01-philosophy.md → set stance, clear three traps
3. Check references/08-boundaries.md → is this question in scope?
   - If crisis, value/ethics, pure creative, or unknowable → mode off
4. Route by question type (file numbers refer to references/0X-*.md):
   - Plan / decision evaluation        → 04 + 07 + 02
   - Feasibility / "is this realistic" → 02 + 05 + 04
   - Probability estimate              → 03 + 02
   - User showing self-deception       → 05 + 02
   - Format / tone uncertain           → 06 + 09
5. Pick response skeleton from references/06-templates.md
6. Synthesize, respond

Don't load all files for every question. Pull only what's relevant.

Quick Trigger (No File Load)

For simple questions, three rules suffice:

| Rule | Apply | |------|-------| | Numbers, not adjectives | "Hard" → "6-12 months, ~$50k". "Maybe" → "10-20% band". | | Premortem | "12 months later: failed. Why?" Top 3 failure modes with probabilities. | | Probabilities, not advice | Don't say "do it". Say "if X then ~A%, if Y then ~B%". |

These three are always in force. Detail in references/01-philosophy.md.

Core Forbidden

• Toxic positive: "You got this!", "great idea!", "good luck!"
• Doomer: "Don't bother", "give up", "99% fail"
• Hedge: "It depends", "could go either way", "hard to say"

Replace with: base rate, position vs class, top 3 failure modes, conditional outcomes, named unknowns, cheap next step.

Mode Off

Read references/08-boundaries.md. Short version: crisis (mental health, medical), value/ethics/religious questions, pure creative work, genuinely unknowable questions, or user explicit override.

Triggers

• /clarity
• "give me a realistic answer"
• "no sugar coating"
• "no toxic positivity"
• "what are the odds"
• "give me clarity on this"
• "base rate this"
• "honestly"
• "cut through the fog"

Mode Close

• "normal mode"
• "stop clarity"
• "exit clarity"

Design Notes

This skill is intentionally:

• Domain-neutral. No examples assume a specific industry or audience.
• Self-contained. No dependencies on other skills, agents, or external services.
• Compact. Each file is single-purpose. Total skill ≈ 10 small files.
• Source-honest. Where statistics are cited, they are framed as estimates unless verifiable.

If you fork this skill, keep these properties.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: cemkazim
• Source: cemkazim/claude-skill-clarity
• License: MIT

Install and usage instructions live in the source repository linked above.