Turn a fork-in-the-road decision into a computed expected-value tree — settle or sue, launch or wait, fix or replace — rolled back by the bundled script, with the break-even probability where the answer flips. Use when asked should we settle or go to trial, build a decision tree, what probability makes this worth it, or compare options under uncertainty. Produces the structured tree, the rollback with the best choice at every fork, the break-even probabilities, and the honest list of what the numbers leave out. Decision support, not advice — the probabilities are yours.
“Should we settle?”“Build a decision tree.”“What probability makes this worth it?”“Compare options under uncertainty.”
What to give it
▸The choices — the real options on the table, including the do-nothing one
▸The uncertainties — what could happen under each choice, and the requester's honest probability for each (pushing back on false precision is part of the job)
▸The payoffs and costs — the money (or a stated proxy) at each end point, and what each path costs to walk: fees, time priced honestly, deposits
▸The stakes context — one-shot or repeatable, and whether the worst branch is survivable — because expected value is the right tool for repeatable bets and needs a caveat for ruinous one-shots
✅ The bar it holds itself to
Every skill in this library self-verifies — these are this skill's own quality checks, straight from its definition.
✓The tree was read back and corrected before anything was computed
✓Every path's costs are on the path, not forgotten at the leaves
✓The break-even scan appears and is compared against the requester's stated belief
✓Values were stressed, not just probabilities
✓The leaves-out list names ruin risk explicitly when the worst branch is severe
✓The recommendation states robustness, not just the EV winner
⚠️ What it refuses to do
**Computing the mis-drawn tree.** Ten minutes of extraction beats any amount of arithmetic on the wrong structure.
**False precision in probabilities.** "About 60%" is honest; "62.5%" from nowhere is decoration — the break-even scan is the cure, since it shows whether the difference even matters.
**EV-maximising a ruinous one-shot.** The tool's cleanest failure mode; the caveat is mandatory, not optional.
**Hiding the arithmetic.** The script prints every node's EV because a recommendation nobody can check convinces nobody who matters.
**Letting the tree end the conversation.** It ends the *circular* part; the values conversation it surfaces is the productive one.
Install
npx pm-claude-skills add --agent claude # or codex · cursor · gemini · hermes
# or one-line MCP (every skill, any client):
claude mcp add pm-skills -- npx -y -p pm-claude-skills pm-claude-skills-mcp