⏳ This skill is pending AI review.
Scores will appear once the review pipeline completes.
allocating-effort
Splits scarce time, budget or traffic across competing products, channels, campaigns or variants by treating them as a multi-armed bandit - Thompson sampling over beta posteriors, with optimism for anything not yet tried enough to judge. Use when deciding what to work on next, how to divide a marketing budget, which product to prioritise, whether to kill something that is underperforming, or when picking a winner among test variants.
Choose how to use this skill
You do not need every option. Choose the path your AI client supports. The stable page stays the same; versioned files are immutable.
1. Native installer
This listing has no registered native installer command. Use the complete package or source fallback below, depending on what your client supports.
Do not guess an installer command or replace an existing version without reviewing the diff.
2. Complete package recommended
Download the ZIP when available. It includes SKILL.md plus the references, security notes and version metadata.
No complete ProSkills package is published for this listing yet.3. Prompt-only
Copy the prompt above when the agent can read the stable page or when you want to adopt the workflow without installing a skill.
Need only the instruction file?
Download SKILL.md only if your client requires a single file. The complete ZIP is safer for a full installation because it preserves the references and release context.
No path installs or executes anything by itself. Your agent still needs access to the project files. Before updating, compare the installed version and review the diff.
// RATINGS
// README
Your agent already gives confident business advice. It will tell you to run the A/B test, match the competitor's price cut, and kill the product with no conversions. Sometimes that advice is right. You cannot tell which times, because there is no number under it.
These eight skills put the number under it.
> The team wants three weeks to build a churn prediction model.
It flags at-risk accounts with 92% accuracy. Worth it?
Value of information: zero. Not low. Zero.
You already send the retention email to everyone in the at-risk segment, and the email costs nothing, so emailing everyone is already the optimal action. There is no prediction the model could return that changes what you do, and an observation that cannot change the action has no value however accurate it is. The 92% is doing rhetorical work, not decision work.
This becomes worth building the day the retention action has a real per-customer cost, because then you have to choose who to spend it on. Test that first: it is cheaper than three weeks of engineering.
That is the whole idea. Eight situations you actually face, eight skills, a calculator behind each one.
What it is
Markdown skills that work with any agent that can read files: Claude Code, Claude Cowork, Codex, Cursor, Windsurf, Cline, Aider. Each skill is backed by a zero-dependency Node calculator, so beta posteriors and expected utilities get computed rather than guessed in prose.
The source is Algorithms for Decision Making by Mykel Kochenderfer, Tim Wheeler and Kyle Wray (MIT Press, 2022): 700 pages on acting under uncertainty, written for collision-avoidance systems and Mars rovers. The mathematics is aimed at aircraft. The problem structure is the one an operator faces every week.
references/book-map.md maps every technique chapter to where it
lands, including an honest list of what does not transfer and a suggested reading order if
you only read six chapters.
What it is not
- Not a forecasting or BI tool. It does no data collection and connects to nothing. You bring the numbers; it tells you what follows from them.
- Not an oracle. Several of the calculators depend on utilities and priors you supply, and they say so in their own output. The value is that those inputs become visible and arguable instead of buried in someone's prose.
- Not a replacement for judgement. Two of the eight skills exist mainly to tell you that a question is a values decision, or that you should stop analysing and ship.
- Not a port of the book's machinery. Policy gradients, neural value approximation and
exact POMDP solvers are in the book and deliberately not here.
references/book-map.mdsays which parts do not transfer and why.
Install
Paste into your agent:
Install all the skills from the github repo
https://github.com/romainsimon/skills-for-decision-making
Then read references/book-map.md and tell me which of my open decisions each one covers.
Or clone it:
git clone https://github.com/romainsimon/skills-for-decision-making
cd skills-for-decision-making && npm run check
Zero dependencies. Node 18 or later.
To install one skill rather than all eight, copy that skill's directory plus scripts/,
lib/ and examples/, which it reaches by symlink.
If your installer fetches files one by one through the GitHub API, it will turn those
symlinks into small text files and the calculators will not run. Clone the repo instead,
or copy the three shared directories in by hand. npm run check catches this: the
validator fails when a skill's scripts entry is not a symlink.
The eight skills
| Skill | Fires when | Gives you |
|---|---|---|
framing-decisions | "should we do A or B", a stalled debate, before any decision memo | Explicit actions, a named unknown, a sourced prior, one utility scale, and a ranked answer with the margin over second place |
valuing-information | someone proposes a test, survey, study, dashboard or consultant | What that information is worth before you buy it, which is zero whenever no result would change the action |
allocating-effort | "what should I work on", budget splits, "should we kill this" | A defensible split across products or channels, and a refusal to cut anything that has not been tried enough times to judge |
planning-horizons | roadmaps, backlog triage, short-term versus long-term arguments | A discount rate derived from runway, the depth past which planning stops paying, and a backlog pruned by arithmetic |
stress-testing-plans | before a launch, price change or migration; premortems | Whether the ranking survives your assumptions, the most likely failure path, and the trade-off you are actually making |
tracking-beliefs | a metric moved and someone wants to act | Whether the move is bigger than noise explains, and which explanation the evidence supports |
reading-rivals | a competitor moved, or you are considering a price change | What they optimise, what they do next, and whether the exchange of responses ends somewhere you want to be |
learning-from-outcomes | retrospectives, postmortems, monthly and quarterly reviews | A calibration score for your own forecasts, credit assigned across time, and changed numbers instead of lessons |
Each one is named for the situation, not the algorithm. Nobody wakes up wanting Thompson sampling.
Ask it things
> We have 90k in the bank, burn 30k a quarter, and I cannot decide between shipping
the paid tier and doubling down on SEO.
> Split 40 hours a week across my five side projects. Product D has 14 signups and
zero conversions, I want to kill it.
> Traffic dropped 30% last week. Everyone says it is a Google update.
> Our competitor just cut prices 30%. My instinct is to match so we do not look weak.
> The paywall made us 4900 a month but organic signups fell from 3100 to 900.
Was it a mistake? Give me one number that settles it.
> Revenue is up 20% and everyone is crediting the paid tier we launched this quarter.
> Leadership wants a three-year plan. We replan monthly. Push back or comply?
> Score my last twelve predictions. Am I actually any good at this?
Some of the answers are "you are asking the wrong question", "that is a values decision and no analysis can make it for you", and "stop analysing and ship". Those are the ones that pay for the repo.
The calculators
Agents are unreliable at beta quantiles, UCB bonuses and expected utility, so none of that is left to prose. One CLI, seventeen commands, no dependencies:
node scripts/calc.js voi study.json # is this research worth doing
node scripts/calc.js meu decision.json # rank
// HOW IT'S BUILT
KEY FILES