decision-journal
GitHub用于记录决策过程以对抗记忆偏差的技能。通过事前预注册推理、概率和证伪条件，并在事后对比结果与预期，严格区分过程质量与运气。旨在通过复盘识别个人思维模式偏差，提升判断力并积累高质量决策经验。
Trigger Scenarios
Install
npx skills add mohitagw15856/pm-claude-skills --skill decision-journal -g -y
SKILL.md
Frontmatter
{
"name": "decision-journal",
"homepage": "https:\/\/mohitagw15856.github.io\/pm-claude-skills\/skill\/decision-journal.html",
"metadata": {
"openclaw": {
"emoji": "⭐"
}
},
"description": "Record decisions the way good judgment compounds — the reasoning, the alternatives, the probabilities, and what would change your mind, written down BEFORE the outcome arrives, then reviewed against reality. Use when asked help me think through this decision, start a decision journal, review my past decision, or why do I keep making the same mistake. Produces the pre-registered decision entry, the review-date trigger, and the outcome review that separates bad luck from bad process."
}
Decision Journal Skill
Memory is a defense attorney: after the outcome arrives, everyone remembers having known it all along, and the reasoning that actually drove the choice is quietly rewritten. A decision journal defeats that — the reasoning gets pre-registered before reality votes, so the review can ask the only question that improves judgment: was the process good, given what was knowable then? Good decisions lose sometimes; bad ones win sometimes; only the written record can tell you which you're making. This is the library's 600th skill because it's the one all the others compound through.
What This Skill Produces
- The entry — situation, the real options, the choice, the reasoning, explicit probabilities, and the pre-registered "I'm wrong if…" markers
- The review trigger — a date and the observable that will have resolved by then
- The outcome review — process vs. luck separated, the delta between expected and actual, and the transferable lesson
- The pattern read — across entries: the recurring biases this specific person's journal reveals
Required Inputs
Ask for these if not provided:
- The decision — what's actually being chosen, and by when; "should I…" questions get reframed into the options actually on the table
- The honest state — what's known, what's guessed, how they feel (mood is data: tired-angry-rushed decisions deserve their own flag in the entry)
- For reviews: the original entry and what actually happened — the review is against the entry, never against memory
Framework: The Pre-Registration Rules
- Write the base case and the alternatives, fairly: the entry states each real option with its strongest case — a journal of straw men reviews into nothing. "Do nothing" is always listed; it's the option most decisions are actually competing against.
- Probabilities, in numbers: "I think this works" becomes "70% this reaches X by [date]." Numbers feel false and are — but a written 70% can be scored later, and "probably" cannot. Calibration only ever comes from this discomfort.
- Pre-register the falsifiers: "I'm wrong if [observable] by [date]" — written now, while it's cheap. This is the line that makes the future review honest and the mid-course correction possible without ego litigation.
- Process vs. outcome, ruthlessly separated at review: four quadrants — good process/good outcome (skill), good process/bad outcome (variance — change nothing), bad process/good outcome (the most dangerous quadrant: luck teaching a bad lesson), bad process/bad outcome (tuition — extract the lesson). The review names the quadrant before discussing feelings.
- Patterns beat entries: every ~10 reviews, read across: systematically overoptimistic on timelines? Underweighting exit costs? Deciding worst when rushed? The journal's compounding value is the personal bias list no generic advice can give.
Output Format
Decision Entry: [title] — [date]
Deciding: [the actual choice + deadline] · State: [known / guessed / mood flag]
Options (each at its strongest)
- … 2. … 3. Do nothing: …
The Call
Choice: [option] · Because: [the 2–3 load-bearing reasons] Expectations: [X% that (observable) by (date); …] I'm wrong if: [observable + date] · Review on: [date]
(At review:)
Outcome Review — [date]
What happened: … vs. expected: … Quadrant: [skill / variance / lucky / tuition] — [why] Transferable lesson: [one sentence, about process not this event]
Quality Checks
- Every option including do-nothing gets its strongest case
- At least one expectation carries a number and a date
- The falsifier is observable, not a feeling
- Reviews compare against the written entry, never memory
- The quadrant is named before the lesson is drawn
Anti-Patterns
- Do not journal after deciding-in-your-heart and call it deliberation — the entry's value is being written while genuinely open
- Do not grade decisions by outcome alone — the lucky quadrant is where bad habits get reinforced
- Do not write unfalsifiable expectations ("this will probably work out") — they review into nothing
- Do not skip small decisions categorically — the journal's patterns come fastest from frequent entries
- Do not let the review become self-flagellation or victory laps — one quadrant, one lesson, close the entry
Version History
- 54fad50 Current 2026-07-19 12:16
