pm-prioritize — Rank requirements with RICE / ICE / MoSCoW / Kano

Part of the Personal Corp framework — running a one-person business through AI agents.

Rank a list of requirements using a structured framework. A built-in decision tree picks the right framework based on data availability and decision context. Output is transparent and traceable, so a team can argue with the scores instead of the recommendation.

Inputs

Field	Required	Notes
Requirement list	yes	Name + brief description; ≥ 3 items. Can take a pain-point list from /pm-feedback or a feature list from /pm-prd
Framework	no	RICE / ICE / MoSCoW / Kano; auto-recommended if not given
Business goal	no	Current focus (growth / retention / revenue / efficiency); affects weighting
Resource constraint	no	Available dev capacity (person-days or Story Points)

Optional config

Most of the skill works out-of-box. If you want stable defaults across runs, add an ## Prioritize Config section to your project's CLAUDE.md:

## Prioritize Config

### Default framework (optional)
If unset, the skill auto-recommends per the decision table below.
- default_framework: RICE | ICE | MoSCoW | Kano

### Default resource constraint (optional)
Used in the Sprint allocation step. Skip if you'd rather state it per run.
- sprint_capacity: 20 person-days per Sprint

### Backlog source (optional)
Where the skill should fetch the requirement list from when you don't paste one.
- backlog_source: gh-issues  # gh-issues | github-project | tasks-file | paste
- gh_owner: your-github-handle
- gh_repo: your-main-repo
- gh_label: backlog
- tasks_file: docs/backlog.md

When a config field is set, the skill uses it silently. When unset, the skill asks (see "When input is incomplete").

Research commands (auto-discovery)

If the user points at a backlog source instead of pasting items, the skill can pull the list itself:

# GitHub issues by label
gh issue list -R $OWNER/$REPO --label $LABEL --state open \
  --json number,title,body --limit 100

# GitHub Project items
gh project item-list $PROJECT_ID --owner $OWNER --format json

# Local backlog file
cat $TASKS_FILE

Step 1 — Pick a framework

If unspecified, recommend per this decision table:

Condition	Recommended	Why
Have user-impact data per item (DAU, conversion), trustworthy	RICE	Most quantitative, traceable
Have intuition but no precise data	ICE	Quick scoring, tolerates subjectivity
Need 4-bucket alignment fast (e.g. team meeting)	MoSCoW	Forces "must" / "won't" consensus
Need to understand requirement nature, plan features	Kano	Identifies delight features

Framework comparison:

Framework	Use case	Strength	Limit	Time
RICE	Data-supported quarterly planning	Most objective, comparable	Depends on data quality	Medium
ICE	Fast decisions, brainstorming	Simple, fast	Highly subjective	Low
MoSCoW	Release planning, stakeholder alignment	Forces consensus	Easy to put everything in Must	Low
Kano	Feature planning, satisfaction research	Identifies delighters	Needs user research data	High

Step 2 — Score requirements

RICE (default)

Dimension	Meaning	Scoring	Common error
Reach	Users impacted in one cycle	Concrete number ("5000 users/month")	"All users theoretically" as Reach
Impact	Per-user impact magnitude	3 = massive, 2 = high, 1 = medium, 0.5 = low, 0.25 = minimal	Everything gets 3
Confidence	Confidence in the estimate	100% = data, 80% = indirect evidence, 50% = gut	100% with no data
Effort	Total person-months across all roles	Includes design + dev + QA + integration	Counting only dev

RICE Score = (R × I × C) / E — higher = higher priority.

Calibration mechanism:

• Score the same dimension across all items first (all R, then all I) — avoids per-item anchoring bias
• R calibration: pick a baseline ("login: affects 100% of users"), score others relative
• I calibration: ≤ 50% of items can score 3 — forces differentiation
• E calibration: must include design (20%) + dev (50%) + QA (20%) + integration (10%)

ICE (fast)

Score 1-10 on each dimension. ICE Score = I × C × E / 10.

Dimension	Scoring
Impact	1 = trivial, 5 = medium, 10 = transformational
Confidence	1 = pure guess, 5 = indirect evidence, 10 = A/B test data
Ease	1 = very hard (> 3 months), 5 = medium (2-4 weeks), 10 = trivial (< 1 day)

MoSCoW

Bucket	Definition	Suggested share
Must Have	Without it, can't ship; users can't use core feature	≤ 60%
Should Have	Important but has workaround; one-Sprint delay non-fatal	~ 20%
Could Have	Nice-to-have; better with, fine without	~ 10%
Won't Have (this time)	Explicitly out of scope; possibly later	~ 10%

Common trap: everything ends up Must Have. Counter: cap Must Have at 60%, force trade-offs.

Kano

Type	Trait	Detection	Strategy
Must-be	Absence → dissatisfaction; presence → taken for granted	Users don't ask for it but rage when missing	Reach passing grade, don't over-invest
One-dimensional	More = more satisfaction (linear)	Users actively request	Core competitive area, top-tier execution
Attractive	Absence → no dissatisfaction; presence → delight	Unexpected, evokes "wow"	Differentiator (decays to one-dimensional over years)
Indifferent	Doesn't matter either way	No user reaction	Don't invest
Reverse	Presence reduces satisfaction	Adds complexity, annoys users	Remove immediately

Kano decay: today's Attractive feature becomes One-dimensional, then Must-be over 2-3 years (e.g. fingerprint unlock). Continuously create new delighters.

Step 3 — Output ranking

RICE results table:

Rank	Requirement	R	I	C	E	RICE Score	Recommendation
1	{name}	{n}	{0.25-3}	{50-100%}	{pm}	{score}	This cycle

Impact × Effort 2×2:

Quadrant	Impact	Effort	Strategy	Items
Quick Wins	High	Low	Do first	{list}
Strategic	High	High	Plan carefully	{list}
Fill-ins	Low	Low	When idle	{list}
Avoid	Low	High	Don't do	{list}

Sprint allocation:

• Allocate per sprint_capacity (config) or stated resource constraint
• Quick Wins fill first; Strategic by RICE Score
• Reserve 10-20% per Sprint for unexpected work

Step 4 — Decision log

• Core trade-offs: why A over B this cycle
• Disputed items: which ranks may be contested, and why
• Confidence flags: which scores have C < 80% — propose validation experiments
• Next-cycle candidates: Won't-Have items most likely to promote next cycle

Quality bar

1. Every score has a one-sentence rationale
2. Effort includes design + dev + QA + integration
3. C < 80% items get "validate via small experiment" tag
4. Must Have ≤ 60% of total
5. Calibration mechanism applied to avoid anchoring

Red lines

1. Not a decision-maker — output is a recommendation; the final call is the team's
2. No hidden assumptions — every score's assumption is explicit
3. Never ignore Effort — no "must do" recommendation based on Impact alone

When input is incomplete

• No backlog source provided → ask: "Where is the list — paste, file path, or a GitHub issues filter (owner/repo + label or milestone)?"
• No business goal → ask: "Current focus this cycle — growth / retention / revenue / efficiency? Optional, but it tightens the recommendation."
• No resource constraint → ask: "Available capacity for the next cycle — person-days or Story Points? Optional, but needed for Sprint allocation."
• Framework unset → don't ask. Auto-recommend per the Step 1 decision table and propose it with a one-sentence rationale; user confirms or overrides.
• < 3 requirements → still rank, but flag "sample too small, add more for stability"
• No data at all → switch to ICE or MoSCoW; tag "qualitative ranking due to missing quantitative data"

Related skills

• weekly-planning — uses the ranked backlog from this skill to pick weekly OKRs / outcomes. Prioritization feeds OKR selection, not replaces it.
• weekly-retro — feeds the next backlog with retro findings and carry-over items
• /pm-user-stories — top-priority requirements → break into Stories
• /pm-prd — Must-Have requirements → write PRDs