| name | decision-log |
| description | Track and structure human decisions made during AI-assisted work. Use when the user makes a choice between AI-generated options, picks a technical direction, or resolves any ambiguity where human judgment matters. Triggers on "log this decision", "record my choice", "why did I choose", "decision log", "track decision".
|
| when_to_use | Any time the user makes a significant decision during development, especially when choosing between AI-suggested alternatives. Also useful for retrospective logging of past decisions and for building a portfolio of human judgment.
|
| allowed-tools | Read Write Edit Bash |
Decision Log
Structured recording of human decisions made during AI-assisted development.
Each entry captures what AI recommended, what you chose, and why — the judgment
that AI cannot replicate.
Quick start
/decision-log "Chose PostgreSQL over MongoDB for the user data store"
/decision-log # Interactive: asks what you decided
/decision-log --review # Review and reflect on past decisions
Why log decisions?
Three reasons, in order of importance:
-
Build judgment. Reviewing past decisions (and their outcomes) is how you
develop product intuition. Six months from now, you'll see which calls were
right and which were wrong — and why.
-
Prove judgment. In interviews or reviews, "I used AI to build this" sounds
like you're a middleman. "Here are the 12 decisions I made that AI couldn't,
and why" shows irreplaceable value.
-
Improve process. If most decisions are "AI said X, I went with X", your
workflow isn't challenging you enough. If most are overrides, your AI setup
may need tuning.
Logging a decision
When the user describes a decision (or invokes /decision-log), structure it:
### Decision: [short title]
**Date:** [auto]
**Context:** [what was being decided]
**Option A:** [what AI/source 1 recommended]
**Option B:** [what AI/source 2 recommended, or an alternative considered]
**Chosen:** [A / B / hybrid / something else entirely]
**Reasoning:** [why — the human context, intuition, or constraint that drove this]
**What AI couldn't weigh:** [the factor that required human judgment]
**Confidence:** High / Medium / Low
**Revisit by:** [optional: date to check if this was the right call]
Append to decision-log.md in the project root. Create the file if it doesn't exist.
Entry types
Mark each entry with a type tag so they can be filtered later:
[PRODUCT] — product direction, feature scope, user prioritization[TECHNICAL] — stack choice, architecture, data model[STRATEGY] — market positioning, pricing, go-to-market[PROCESS] — workflow, tool choice, team structure[REVIEW] — AI-assisted review finding (not a human decision — labeled separately)
The [REVIEW] type is for recording AI evaluation results (like blind review
findings or user simulation outputs). These are valuable records but should be
clearly distinguished from human decisions.
Interactive mode
When invoked without arguments, ask:
- "What did you just decide?"
- "What were the alternatives?"
- "Why did you go this direction?"
- "What's the one thing AI couldn't factor in?"
Format the answers into the standard template and append.
Review mode (--review)
Read the existing decision-log.md and provide:
- Decision count by type
- Override rate — how often did the user diverge from AI recommendations?
- Confidence distribution — are most decisions High or Low confidence?
- Patterns — recurring themes in "what AI couldn't weigh"
- Revisit flags — any decisions past their revisit date
Present as a brief summary, then ask: "Any of these decisions you'd make
differently now?"
Output files
| File | Purpose |
|---|
decision-log.md | Append-only decision record |
