Topic leaf H1 for ai-prompt-template-builder now names prompt builders/generators and every scene; rules and lint block “which X page” titles. | Dev Logs

Date: 2026-09-02 15:30 Summary: Topic leaf H1 for ai-prompt-template-builder now names prompt builders/generators and every scene; rules and lint block “which X page” titles. Visibility: people

[question] http://localhost:8787/en/topics/ai-prompt-template-builder/ 页面title 机器翻译过于明显，应该把场景都说清楚，另外不是 prompt page，而是prompt builder/generator， 分析产生这样问题的原因，并在规则中避免其他工具和topic的内容生成中出现相同类似的问题

[try to solve]

Context

The English topic leaf at /en/topics/ai-prompt-template-builder/ rendered H1/title as “Writing, Midjourney, or iOS agent: which prompt page”. That reads like a calque of the Chinese “写作、Midjourney，还是 iOS 智能体：选哪张提示词页”: a site-navigation question, a generic “page” product noun, and only three of nine scene tools.

Process

1. Traced the live string to scripts/site/topic-ig-copy-*.mjs (flattened by build:topic-ig into i18n-topic-ig.mjs, merged last in t() so it overrides the older hub title “AI prompt template builder & generator tools”).
2. Compared with the same-day topic IG pass (dev-logs/2026-09/2026-09-02-11-17-topic-leaf-ig-decision-tables.md), which instructed “H1 is a decision question” and used an A, B, or C: which X … skeleton across 18 leaves.
3. Confirmed localization globs did not include topic-ig-copy-*.mjs, so agents writing topic copy were not loaded with tool-i18n-localization.mdc.
4. Rewrote all ten locale H1s to name builder/generator (local search wording) and list the independent scenes (writing, Midjourney, sketch, film, short drama, product design, Android, iOS). Also fixed the same chrome-noun bug on English/Chinese text-compare (which compare page / 选哪张对比页).
5. Encoded the failure mode in rules (tool-i18n-localization.mdc, tool-i18n-seo.mdc, seo-html-defaults.mdc, AGENTS.md) and in lint:seo via scripts/lib/topic-title-heuristics.mjs. Regenerated IG i18n, rebuilt topic HTML, gzipped _pages. lint:seo reports Topic leaf title validator: OK.

Root cause / analysis

Three stacked mistakes, not a single bad string:

1. Wrong H1 template. The topic IG pass treated every leaf H1 as a “which X first” decision question. For BMI vs TDEE that can still name the user’s job. For this cluster it became “which page”—chrome, not a search query. Users look for prompt builder / prompt generator.
2. Cherry-picked scenes to fit “A, B, or C”. Nine tools (generic template + writing + Midjourney + sketch + film + short drama + product design + Android + iOS) were reduced to Writing / Midjourney / iOS so the title stayed short and parallel. That hides the actual scene set.
3. Isomorphic translation + missing rule load. English tracks the Chinese “选哪张…页” skeleton (página de prompt, Prompt-Seite, プロンプトページ, Arabic even drifted to صفحة أمر “command page”). Topic copy files were outside localization globs, so the existing “rewrite, don’t translate” rule never fired on this pass.

lint:seo previously had no topic-H1 heuristic, so the bad titles shipped green.

Solution

Live English H1 / <title> (before brand):
AI prompt builders and generators for writing, Midjourney, sketch, film, short drama, product design, Android, and iOS

Chinese and the other eight locales follow the same facts (product type + full scene list) with local wording, not a locked English frame.

Rules (durable): Title/H1 must use the search product type, never page/页 as the product name; scene clusters must name the independent scenes (or a grouping that covers the whole set); topic H1 is not required to be a decision question; do not calque “选哪张 X 页”. AGENTS.md now tells agents to load localization rules when editing /topics copy.

Gate: checkTopicLeafTitle runs in build:topic-ig (before flatten) and validateTopicLeafTitles in lint:seo (on generated keys). Prompt cluster titles must match local builder/generator words and hit at least six of eight scene tokens.

Notes / boundaries

• Other topic H1s that ask a user-job question (“BMI, TDEE, or macros: which health number first”) were left in place; only chrome-noun titles (prompt page, compare page) were rewritten.
• Nav short label topic_ai_prompt_template_builder remains “AI prompt tools”.
• Local wrangler on :8787 did not respond during this pass (probe hung). Disk HTML is updated under public/_pages/{lang}/topics/ai-prompt-template-builder.html; seeing it in the browser needs a healthy start:dev and local R2 seed (upload:r2:local if the bucket still has the old gzip).
• Did not rewrite remaining “decision question” H1s across all 18 topics; the new rule applies to future tool and topic copy.

[actions]

• Updated tool-i18n-localization.mdc, tool-i18n-seo.mdc, seo-html-defaults.mdc, project-core.mdc, AGENTS.md, .github/copilot-instructions.md.
• Added scripts/lib/topic-title-heuristics.mjs and scripts/validate-topic-leaf-titles.mjs; wired into build-topic-ig.mjs and lint:seo.
• Rewrote prompt-cluster titles in topic-ig-copy-en.mjs / zh / rest locales; English/Chinese text-compare titles; regenerated i18n-topic-ig.mjs; rebuilt and gzipped topic _pages HTML.