skills.rest
Raynos
RaynosCOMMUNITY
Explore all skills by this author →
𝕏@RaynosGitHub

distill-taste

Triages human feedback corpus items into a capped taste standard document.

1|Updated Jun 29, 2026
Search Tags:#design principles#game design#playtesting#feedback triage#taste standard#doc budget#prediction test
One-click install
npx skills add https://github.com/Raynos/kami-kakushi --skill distill-taste-raynos
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: distill-taste
Source: https://github.com/Raynos/kami-kakushi/tree/main/.claude/skills/distill-taste
Command: npx skills add https://github.com/Raynos/kami-kakushi --skill distill-taste-raynos

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Playtest and feedback sessions accumulate raw notes that risk bloating or diluting a design taste standard. This Skill distills a feedback corpus into the locked taste standard (docs/living/taste.md) without growing it past its line budget, keeping the standard sharp and falsifiable. ## Core Features & Use Cases - Five-move triage: Classifies every feedback item as repeat-offense, new evidence, genuinely new taste, wrong-home taste, or one-off verdict, preserving verbatim quotes and F-numbers. - Budget and coverage verification: Runs verify-doc-budgets, checks every corpus F-number is covered, and prediction-tests substantial changes with blind agents. - Human-locked top layer: Proposes but never applies changes to values, touchstones, or references without explicit human approval recorded as an ADR. - Use Case: After a playtest session produces project/feedback-human notes, run /distill-taste to fold the verdicts into taste.md, tag distilled items in the corpus, and journal the repeat-rate. ## Quick Start Run /distill-taste on the newest feedback corpus file to triage every item into the taste standard and verify budgets and coverage.

Frequently Asked Questions about distill-taste

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I distill playtest feedback into a design taste standard?▼

Run /distill-taste with the feedback corpus file. It reads the current taste standard fresh, triages every item into one of five moves, edits in place without appending, then verifies budgets, coverage, and prediction tests before committing.

What are the five triage moves for feedback items?▼

Items are classified as repeat-offense (strengthen enforcement, not prose), new evidence for an existing principle, genuinely new taste requiring a new principle, wrong-home taste routed to other docs, or one-off verdicts that stay corpus-only.

Can the skill change the core values in the taste standard?▼

No. The four values, touchstones, and references are human-locked. The skill proposes edits but requires explicit human approval via a question flow, recorded as an ADR amending ADR-126, before any top-layer change is applied.

How is distillation quality verified?▼

Verification runs the doc-budget script (taste.md at 150 lines or fewer), set-diffs F-numbers for full corpus coverage, and prediction-tests substantial changes with two blind agents judging disguised cases. A convergent miss triggers a wording fix and re-run.

When should I not add a new principle to the standard?▼

Do not add a principle for repeat offenses already covered by existing rules, for one-off fixes with no standing rule behind them, or when sharpening an existing principle suffices. Default to sharpening over multiplying to prevent rule-count creep.

Related Skills

Explore More Product & Management Skills→
wanghaishengwanghaisheng
10

playtest-report

Generate structured Markdown playtest report templates and analyze feedback.

Community
Intermediate
maragudkmaragudk
27

diary

Automate structured implementation diary entries with frontmatter and step-wise guidance.

Official
Intermediate
ZenkakuHiraganaZenkakuHiragana

empirical-prompt-tuning

Run blinded subagents against a frozen evaluation design to improve agent-facing instructions.

Community
Advanced
austintheriotaustintheriot
1

one-by-one

Guide sequential item-by-item decisions on lists with end-only synthesis.

Community
Intermediate