Problem
The recommendation-gradient format presents pros and cons inline, separated by ;. When options carry multiple items or longer text, the inline format is hard to parse — option boundaries blur and individual pros/cons run together.
Context
The format is defined in packages/agents/content/skills/_data/recommendation-gradient.md and modeled in two other places:
packages/agents/content/skills/_data/design-priorities.md — convenience-led / correctness-led ranking examples
packages/agents/content/skills/design-and-plan/SKILL.md — inline reminder bullet (line 71)
Design decisions:
- Always put each pro/con on its own line, even when an option has only a single item. One rule, predictable output, no per-case judgement.
- Indent each item line by 3 spaces, aligning with the option title's first character after
1. .
- Keep
➕ / ➖. No Unicode pair offers reliable color in Markdown/terminal rendering, and emoji alternatives (✅/❌, 🟢/🔴) carry semantic mismatches or collide with existing markers (the 🟢 recommended next-steps convention).
- Keep punctuation:
; between items, . on the last.
Solution
Update the canonical "Format" section and both example blocks in recommendation-gradient.md to require one pro/con per line with a 3-space indent. Reformat the two ranking examples in design-priorities.md to match. In design-and-plan/SKILL.md, remove the inline example from the recommendation-gradient bullet — the canonical spec is already linked, so duplicating the format invites drift.
The new shape:
1. ■■■ Option title:
➕ first pro;
➕ second pro;
➖ con.
Acceptance criteria
Problem
The recommendation-gradient format presents pros and cons inline, separated by
;. When options carry multiple items or longer text, the inline format is hard to parse — option boundaries blur and individual pros/cons run together.Context
The format is defined in
packages/agents/content/skills/_data/recommendation-gradient.mdand modeled in two other places:packages/agents/content/skills/_data/design-priorities.md— convenience-led / correctness-led ranking examplespackages/agents/content/skills/design-and-plan/SKILL.md— inline reminder bullet (line 71)Design decisions:
1..➕/➖. No Unicode pair offers reliable color in Markdown/terminal rendering, and emoji alternatives (✅/❌,🟢/🔴) carry semantic mismatches or collide with existing markers (the🟢 recommendednext-steps convention).;between items,.on the last.Solution
Update the canonical "Format" section and both example blocks in
recommendation-gradient.mdto require one pro/con per line with a 3-space indent. Reformat the two ranking examples indesign-priorities.mdto match. Indesign-and-plan/SKILL.md, remove the inline example from the recommendation-gradient bullet — the canonical spec is already linked, so duplicating the format invites drift.The new shape:
Acceptance criteria
recommendation-gradient.mdrequires one item per line with 3-space indent, applied even to single-item options.recommendation-gradient.mdreflect the new format.design-priorities.mdreflect the new format.design-and-plan/SKILL.mdno longer contains an inline format example; it refers to the canonical spec only.