agentleFS
calvyntwhcalvyntwh/karu-custom-skills/skills/decision-matrix/SKILL.md

decision-matrix

Skill1 starsChanged 2 months agoMITView on GitHub

Objective trade-off analysis using weighted criteria. Eliminates popularity bias by requiring user-defined priorities. Use when choosing between technologies, architectures, or vendors.

16 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

12 rules. 0 are common, 12 are distinctive, and 0 are very specific to this project.

  • 0

    Common

    Most of the 13 edited versions we read keep these.

  • 12

    Distinctive

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Any Multi-Option Decision: When there are 3+ options and no obvious winner.

    line 17
  2. Distinctive:

    2 options or fewer: Use Pros/Cons instead. Matrix adds overhead without benefit.

    line 21
  3. Distinctive:

    Non-compensatory criteria: If any criterion is an absolute blocker (e.g., "must support IE11"), filter those options first.

    line 23
  4. Distinctive:

    High uncertainty / no data: When you cannot evidence any scores, the matrix produces false precision.

    line 24
  5. Distinctive:

    Already decided: If the user has already chosen, the matrix will confirm rather than inform.

    line 25
  6. Distinctive:

    Ask the user for their Top 3-5 evaluation criteria.

    line 38
  7. Distinctive:

    Before proceeding, ask the user:

    line 47
  8. Distinctive:

    If yes, filter out disqualified options before scoring.

    line 51
  9. Distinctive:

    Ask the user to rate the importance of each criterion.

    line 54
  10. Distinctive:

    Before presenting results, verify:

    line 92
  11. Distinctive:

    Time budget: If analysis exceeds 30 minutes, simplify to binary (Meets/Doesn't meet) or Pros/Cons

    line 102
  12. Distinctive:

    Review: If no entries in 60+ days, check LEARNINGS.md before next use.

    line 146

“Common” is measured: kept by at least 6 in 10 of the 13 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Decision Matrix (The Bias Eliminator)

"When you can measure what you are speaking about... you know something about it." - Lord Kelvin

When to Use

  • Technology Choices: "Should we use React or Vue?"
  • Architecture Decisions: "Monolith vs Microservices?"
  • Vendor Selection: "AWS vs GCP vs Azure?"
  • Any Multi-Option Decision: When there are 3+ options and no obvious winner.

When NOT to Use

  • 2 options or fewer: Use Pros/Cons instead. Matrix adds overhead without benefit.
  • Incommensurable options: Cannot compare "adopt React" vs "build custom framework" if the evaluation criteria cannot be objectively measured.
  • Non-compensatory criteria: If any criterion is an absolute blocker (e.g., "must support IE11"), filter those options first.
  • High uncertainty / no data: When you cannot evidence any scores, the matrix produces false precision.
  • Already decided: If the user has already chosen, the matrix will confirm rather than inform.

The Protocol: Weighted Scoring

0. Pre-Check (Occam's Razor)

  • 2 Options Only? Skip the matrix. Use a simple Pros/Cons list.
  • 3+ Options? Proceed with the full matrix.

1. List Options

Enumerate all viable choices.

  • Example: React, Vue, Svelte, Angular

2. Define Criteria (WITH USER)

Ask the user for their Top 3-5 evaluation criteria.

[!IMPORTANT] You MUST ask the user for criteria. Do not invent criteria based on general knowledge. The user's context defines what matters.

  • Example Criteria: Performance, Learning Curve, Ecosystem Size, Bundle Size, Hiring Pool

3. Check for Conflicts

Before proceeding, ask the user:

  • "Do any of these criteria conflict with each other?" (e.g., "max performance" vs "min bundle size")
  • "Is any criterion non-negotiable — meaning if any option fails it, it's automatically disqualified?"

If yes, filter out disqualified options before scoring.

4. Assign Weights (WITH USER)

Ask the user to rate the importance of each criterion.

WeightMeaning
1Nice to have
2Moderately important
3Very important
4Critical (high weight, but compensable — non-negotiables should be filtered in Step 3 instead)

5. Score Options (with Evidence)

Rate each option against each criterion (1-5 scale).

ScoreMeaningEvidence Required
1PoorMust cite specific data point
3AdequateDefault if uncertain
5ExcellentMust cite specific data point

[!WARNING] No evidence = Score 3. If you cannot cite benchmarks, docs, or user research, default to "Adequate" (3). Do not guess.

You (the agent) can score based on research, but weights come from the user.

6. Calculate Weighted Scores

For each option: Total = Σ (Score × Weight)

Present results as a single Total per option (e.g., "Vue: 40, React: 35, Angular: 31"). Use ranges only when the user explicitly asks for uncertainty estimates.

7. Recommend & Verbalize

Present the matrix and state the winner with reasoning.

  • "Based on your priorities (Performance=Critical, Learning Curve=Important), Vue scores highest because..."

8. Set Review Trigger

Set a 6-month review checkpoint. Log this decision to .learnings/REVIEW.md.

Validation Steps

Before presenting results, verify:

  1. Non-compensatory filter applied: Were hard constraints checked first?
  2. Conflict detection done: Were negatively correlated criteria surfaced?
  3. Evidence anchoring: Did you cite evidence for scores of 1 or 5?
  4. Uncertainty notation: Did you use ranges for high-uncertainty outputs?

Timeboxing

  • Maximum criteria: 7 (beyond this, weights become statistically meaningless)
  • Maximum options: 10 (beyond this, scoring becomes inconsistent)
  • Time budget: If analysis exceeds 30 minutes, simplify to binary (Meets/Doesn't meet) or Pros/Cons

Example

User Goal: Choose a frontend framework for a small team with tight deadlines.

User Criteria & Weights:

CriterionWeight
Learning Curve4
Performance2
Ecosystem3

Agent Scoring:

OptionLearning (×4)Perf (×2)Ecosystem (×3)Total
React3 (12)4 (8)5 (15)35
Vue5 (20)4 (8)4 (12)40 ✅
Angular2 (8)4 (8)5 (15)31

Recommendation: "Vue scores highest (40) primarily due to its excellent Learning Curve score, which you rated as Critical."

Skill Integration

SituationUse Instead/Also
2 optionsrubber-ducking
Need to validate scoresmap-vs-territory
Criteria might be wrongchestertons-fence
Risk of over-analysisoccams-razor
Need to understand second-order effectssecond-order-thinking
Prioritizing a backlogpareto-principle

Self-Improvement Protocol

Log only if prediction was wrong.

## [YYYY-MM-DD] {Brief Description}
**Decision:** {what was chosen}
**Outcome:** {Correct or wrong?}
**Lesson:** {one sentence}

Review: If no entries in 60+ days, check LEARNINGS.md before next use.

Evaluations

Eval 1: Three-Way Technology Choice

Scenario: User must choose between Postgres, MongoDB, Redis for a new project. No existing codebase. Expected: Asks for 3-5 criteria with weights, scores each option, produces weighted recommendation. Pass criteria: Requires user input for criteria/weights, cites evidence for scores of 1 or 5.

Eval 2: Two Options (Should Skip)

Scenario: User asks "React or Vue for a startup?" Expected: Recognizes 2 options, redirects to Pros/Cons, does NOT build full matrix. Pass criteria: Skips matrix, offers simpler alternative.

Eval 3: Non-Compensatory Filter

Scenario: User asks "pick a cloud vendor" with constraint "must support HIPAA." Expected: Identifies HIPAA as non-negotiable, filters vendors BEFORE scoring. Pass criteria: Applies filter first, then matrix only on remaining options.

Resources

  • Detailed Research Notes

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.