本文へ移動
cccskills
無料GitHub で公開

product-ai-feature-scoping-and-evals

Use when the user names an AI feature, an agent, a prompt, retrieval, evals, or model quality work: the scoped spec with a prompt set, a four-rung quality ladder, a scoreboard read over three runs, and the safety gates.

日本語の概要は準備中です。原文の説明を表示しています。

実行スコア 未検証

インストール方法を見る

含まれるファイル(1)

  • SKILL.md2.6 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

AI feature scoping and evals

Use this when the user names an AI feature, an agent, a prompt, retrieval, evals, or model quality work. The output is a scoped spec whose quality claim is measured rather than asserted.

What you need first

The job the AI does, the users, the quality bar, and the cost of being wrong. No quality bar means you propose one from the failure cost and mark it INFERENCE. No eval data means you build the prompt set before any talk about which model to use.

Scope it, then measure it

  1. Say back the job in one line: the input, the output, and who is harmed when it is wrong. High-harm outputs get human review by default — say so early.
  2. Build the prompt set first: twelve to twenty prompts across the real kinds of input the feature will see — happy path, edge, adversarial, and out-of-scope — and mark the top five that must never regress.
  3. Set the quality bar as a ladder with four rungs: nailed it, usable with caveats, wrong but harmless, harmful or fabricated. Every rung needs an example, not an adjective.
  4. Capture before you judge: collect the model's exact outputs first, then grade each against the ladder with exact quotes. Never grade from memory.
  5. Score it on a board: counts by rung, broken down by prompt kind, with a three-run trend rule. One run is noise; three runs is a read.
  6. Spec the trade-offs: quality against latency, cost per task, and safety. Name the red-team prompts and the audit trail the feature keeps. For retrieval-augmented generation, spec what gets retrieved, from where, and what happens when nothing relevant comes back.

Output

The dated eval plan with the prompt set, the ladder, and the board, plus the feature spec it feeds. Offer to rerun on a yes whenever the model or the prompt changes. Point at an existing open-source eval harness for the running of it rather than building one.

When the outputs are not measured

Say the number is not measured and refuse to quote one. Never claim a quality number you did not collect, and never ship a high-harm output without a gate and the owner's yes.

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

コメント

ログインなしで書き込めます。使ってみた感想、質問、動かなかった環境などを気軽にどうぞ。

まだコメントはありません。最初のコメントを書いてみませんか。

送信するとすぐに公開されます。メールアドレスや電話番号などの個人情報は書かないでください。URL は 2 つまでです。

0 / 1000 文字

このフォームは reCAPTCHA で保護されており、Google のプライバシーポリシーと利用規約が適用されます。

同じリポジトリのスキル

Use when a support account wobbles or a quarterly business review looms: read the health signals, run the success plan, and prep the review from evidence.

日本語の概要は準備中です。原文の説明を表示しています。

Significant-Gravitas/skills-catalog22026年10月22日 更新

Reconcile accounts by comparing GL balances to subledgers, bank statements, or third-party data. Use when performing bank reconciliations, GL-to-subledger recs, intercompany reconciliations, or identifying and categorizing reconciling items.

日本語の概要は準備中です。原文の説明を表示しています。

Significant-Gravitas/skills-catalog22026年10月22日 更新

Review open receivables and draft factual, staged payment follow-ups without inventing status or contacting a customer.

日本語の概要は準備中です。原文の説明を表示しています。

Significant-Gravitas/skills-catalog22026年10月22日 更新

Write ad variants that each test one idea, within platform limits and supported claims.

日本語の概要は準備中です。原文の説明を表示しています。

Significant-Gravitas/skills-catalog22026年10月22日 更新

Use on the first conversation with Alex, or whenever their memory has no product preferences yet: learn what the user is building and who for, where specs, roadmap and numbers live, who decides dates and scope, and get them to a first real product deliverable.

日本語の概要は準備中です。原文の説明を表示しています。

Significant-Gravitas/skills-catalog22026年10月22日 更新

Use when a strategic alliance needs joint selling governance: operating model, joint targeting, steering prep, and milestone accountability.

日本語の概要は準備中です。原文の説明を表示しています。

Significant-Gravitas/skills-catalog22026年10月22日 更新

Significant-Gravitas のスキルをすべて見る

このスキルの問題を報告する