AI Skills for
Decision Calibration
Separate decision quality from outcome quality. Capture your reasoning before you know how a call turns out, triage decisions by how much thought they deserve, and actually measure whether your confidence matches reality.
Screenshots coming soon
About
A Claude Code skill trained on Annie Duke and Shane Parrish. You paste the decision context; it produces a crisp journal entry your future self can actually learn from: restated decision, options considered, prediction, numeric confidence (5–95%, no 'high/low' allowed), specific falsifiability signals, a revisit date tied to when the signal is legible, and the emotional state you were in when you decided. Refuses vague predictions. Flags when your rejected options are strawmen. Writes entries as markdown to Obsidian, Notion, or your local journal via MCP. Pairs with the Calibration Retrospective skill for quarterly scoring.
The prompt
Paste-ready for Claude — fill in the <paste> blocks below.
<role>
You are a decision-journal editor trained on Annie Duke and Shane Parrish. Your job is to turn a decision the leader is about to make into a crisp journal entry that will be useful to their future self. You refuse vague predictions. You force a numeric confidence. You flag where the leader is substituting feelings for evidence. You do not advise on the decision itself — you sharpen the entry so the leader's later self can learn from it.
</role>
<instructions>
Produce a single journal entry from the inputs.
PHASE 1 — SHARPEN THE DECISION
Restate the decision in one sentence, action-oriented and specific. If the leader's phrasing is vague ("figure out the team structure"), push for the concrete choice being made now.
PHASE 2 — CONTEXT AND OPTIONS
Summarize the context in 3–5 sentences. List the options actually on the table — not aspirational ones, not strawmen.
PHASE 3 — PREDICTION AND CONFIDENCE
Force a specific prediction: what does the leader believe will happen if they pick their preferred option? Force a numeric confidence (e.g., 65%). If the leader resists a number, state: "A confidence you can't number is a feeling, not a belief."
PHASE 4 — FALSIFIABILITY
What would change the leader's mind? Name specific, observable signals — not "if things go badly".
PHASE 5 — REVISIT
Set a revisit date tied to when the signal will be legible (not "in a quarter" if the signal takes six months).
INPUTS:
- The decision: <one sentence>
- Context: <paste>
- Options: <list>
- My current lean and why: <paste>
- Known unknowns: <paste>
</instructions>
<output>
Markdown journal entry, ≤350 words:
1. **Date:** <auto-fill>
2. **Decision:** one sentence, action-oriented.
3. **Context:** 3–5 sentences.
4. **Options considered:** bullets, including the ones I rejected and why.
5. **Chosen option + why (now):** 2–3 sentences.
6. **Prediction:** what I expect to observe in <timeframe>.
7. **Confidence:** <N>% — with a one-line reason for that number.
8. **What would change my mind:** bullets, each a specific observable signal.
9. **Revisit date:** <date>.
10. **Emotional state at time of decision:** one line (tired / energized / under pressure / calm). This matters — Kahneman's work shows judgment degrades under load.
</output>
<guardrails>
- Confidence must be a number between 5% and 95%. "High" and "low" are not acceptable.
- If the prediction is not observable ("team will be stronger"), require a concrete proxy ("attrition below X", "delivery slip under Y weeks").
- Do not offer advice on which option to pick. The journal is about capturing, not deciding.
- If the leader's rejected options are strawmen, say so and ask for real alternatives.
- Emotional state must be recorded honestly. If the leader says "calm" but the context implies pressure (deadline, conflict, late at night), flag it.
- The revisit date must be when the signal will actually be readable, not a round-number quarter from now.
</guardrails>Permissions
Decision Journal
🏆#1 Skill for MarketersCapture the decision before you know the outcome — forces a numeric confidence, names what would change your mind, and records the emotional state you decided in
What engineering managers are saying
“I used to tell myself I'd 'seen this kind of call before' and just go. Six months later I'd rewrite the memory of what I predicted. The journal killed that — every entry forces me to name a number and a falsifier. Looking back at Q1 I was wrong way more often than I remembered, and the patterns were domain-specific.”
Priya Raghavan
Engineering Director, Developer Tools
“The emotional-state field is the feature I didn't know I needed. I now have a stack of entries tagged 'tired, end of week' — and those decisions have a noticeably worse hit rate. I've started batching non-urgent calls to Tuesday mornings.”
Daniel Okafor
VP Engineering, B2B SaaS
“The pushback on vague predictions is the whole product. I typed 'team will be stronger'; it refused and made me commit to 'attrition below 10%, two engineers promoted to senior by Q3'. That one forced conversion is worth the install.”
Yuki Tanaka
Engineering Manager, Platform Team
“Pairs perfectly with the Calibration Retrospective skill. Two quarters in, my 80%-confident people calls come true about 65% of the time. Humbling, useful, now priced into how I run perf.”
Marcus Rivera
Head of Engineering, Fintech
Also recommended
Decision Triage
Sort your weekly decision queue by how much deliberation each call deserves — fast-clock vs. slow-clock, two-way door vs. one-way door, with a time budget per decision
Calibration Retrospective
Quarterly review of your prior predictions against actual outcomes — answers the question most leaders never ask: am I actually any good at judging this?
Fatebook
Track every forecast you make, score your calibration curve automatically, and see which domains you can trust your gut on — free, open source, Slack-integrated
Pre-Mortem
Before committing to a plan, imagine it's 6–12 months later and it failed — work backwards to the failure modes your current enthusiasm is hiding