AGENT SKILL

tradeoff-matrix

Score options against weighted criteria; surfaces where the numeric ranking disagrees with the gut call and names the tie-breaker.

License 许可
MIT
最近更新
2026-10-29
Stars
★ 0
Forks
0

01 前言 Front Matter

name
tradeoff-matrix
description
Score options against weighted criteria; surfaces where the numeric ranking disagrees with the gut call and names the tie-breaker.

02 资源文件 1

03 SKILL.md 正文

tradeoff-matrix

Pin a decision's criteria + weights + scores. Output is a matrix and an honest discussion of what's not captured by the numbers.

Mode: analytical — judgment inside a fixed frame. Keep the output shape; every verdict cites its evidence; unknowns are marked [TBD] or [VERIFY], never filled in.

When to use

  • Two or more reasonable options for a decision.
  • The team is going in circles because everyone weights criteria differently.
  • A workflow's decision-support step needs a structured comparison.

Inputs

  • The question being decided (one sentence).
  • The candidate options.
  • Optional: candidate criteria + weights from the caller. Weights are the decision-owner's priorities: when they are absent and nobody can answer (unattended run), propose them, label the weight column proposed, and say so in the summary.
  • Scores come from evidence in the inputs; a score you had to guess carries ? (4?).

Procedure

  1. Read Tradeoff-matrix.md.

  2. Slug from the question: a workflow's ${workflow.slug} wins; otherwise lowercase it, replace each run of characters outside a-z0-9 with -, trim - at the ends, and cut to ≤ 60 characters at a -. An existing file at the target path is updated in place (show the diff), never duplicated.

  3. Pick criteria — 3–6 that genuinely differentiate, each with a weight (1–5) and a one-line rationale — and score each option 1–5 per criterion.

  4. Compute with the bundled script — never by hand. It prints the per-cell math (w × s = v), totals, ranking, non-differentiating criteria, and the full sensitivity sweep (every weight ±1 and every score ±1):

    python3 "${POLYMATH_PLUGIN_ROOT:-${CLAUDE_PLUGIN_ROOT}}/skills/tradeoff-matrix/scripts/weigh.py" <<'CSV'
    criterion,weight,A. In-memory,B. Redis,C. Postgres
    Latency (P99),5,5,5,3
    Operational burden,4,5,3,4
    CSV
    

    Re-cut or drop any criterion it lists as non-differentiating, then re-run.

  5. Draft docs/decisions/<slug>-tradeoff.md from the template, pasting the script's table, ranking, and sensitivity lines verbatim:

    • Criteria — table of criterion, weight, rationale.
    • Options × criteria — the script's table.
    • Sensitivity check — the script's lines. Every criterion it says the decision hinges on needs evidence, not opinion; a flip on a GUESSED score means the decision is not ready.
    • Discussion — where the numeric ranking disagrees with the gut call, name the tie-breaker explicitly. Often the "missing criterion" is named here.
    • Recommendation — the option that wins after discussion (which may not be the highest scorer).
    • What would change the answer — name the future condition that would flip the choice.

Quality bar

  • Criteria differentiate the options. Re-cut any criterion that every option scores the same on.
  • Weights have a one-line rationale.
  • Math is shown per cell and every total and sensitivity line came from the script; no "Option B: 7/10" hand-waves.
  • If the recommendation contradicts the numeric ranking, the tie-breaker is named explicitly.
  • The sensitivity check is stated: either "no single ±1 change flips the winner" or the criterion the decision hinges on.

Output

  • File: docs/decisions/<slug>-tradeoff.md.
  • One-line summary listing the chosen option + the tie-breaker (if any).

Anti-patterns to avoid

  • Choosing criteria to make a preferred option win ("retrofitting").
  • Anything scored 5 across all options.
  • Hidden weights inside a single "fit" score.
  • Recommendation that picks the lowest-scoring option without explaining why.

04 同仓库更多 Skills