跳至正文
OpenAgentSkill
浏览已开放 · 目录记录日期 2026-11-21 · 自动更新和账户服务暂时关闭。
设计与 UI

using-humane

Route a design task to the right humane skill, and after any humane skill finishes, name the next one from the declared handoff graph instead of guessing. Reads the `handoffs:` / `accepts:` / `orchestrates:` keys in each skill's frontmatter, so the route is data, not recollection. Use at the start of design work when it is not obvious which skill owns the task, and at the end of any humane skill to find what it hands to. Not for non-design work. Triggers on where do I start, which humane skill, what comes next, what should I run after, humane cycle, design cycle order, "I finished the jtbd interview now what", "какой скилл дальше".

glebis · 已有 AI 审核记录

安装与使用

运行命令前先阅读来源与依赖说明。复制命令不会自动安装或执行。

npx skills add glebis/humane-agentic-design --skill using-humane

技能说明


name: using-humane description: Route a design task to the right humane skill, and after any humane skill finishes, name the next one from the declared handoff graph instead of guessing. Reads the handoffs: / accepts: / orchestrates: keys in each skill's frontmatter, so the route is data, not recollection. Use at the start of design work when it is not obvious which skill owns the task, and at the end of any humane skill to find what it hands to. Not for non-design work. Triggers on where do I start, which humane skill, what comes next, what should I run after, humane cycle, design cycle order, "I finished the jtbd interview now what", "какой скилл дальше".

Using humane

Announce at start: "I'm using the humane:using-humane skill to route this to the skill that owns it."

Scope

This skill routes design work. It does not claim authority over unrelated tasks — a bug fix, a deploy, or a shell question has no humane owner, and routing one here is a false positive. If nothing below matches, say so and move on rather than stretching a design skill over a non-design task.

The two questions it answers

"Where do I start?" — match the intent to an entry point (table below).

"I just finished X, what now?" — do not recall the cycle from memory. Ask the graph:

python3 scripts/graph.py --from layout-rules   # what it hands to, and when
python3 scripts/graph.py --to ux-writing       # who hands in here, and when
python3 scripts/graph.py                       # the whole graph
python3 scripts/graph.py --mermaid             # the graph as a diagram

The script reads the frontmatter of every skill in the plugin. It is stdlib-only and has no dependencies, so it runs anywhere the plugin is checked out.

Entry points

The user saysStart withNot with
"something's missing / nothing works / fresh machine"humane:setupanything else — it diagnoses first
"what should I build", "what are people hiring this for"humane:jtbda review skill; there is nothing to review yet
"mock this up", "show me the flow", "wireframe this"humane:prototypehumane:design-tokens — the structure is not settled yet
"implement this with Astryx", "fit this to our design system"humane:design-frameworkshumane:prototype — use the adapter after the product decision is settled enough to fit
"we need a brand / an identity"humane:brandkithumane:design-tokens — the tokens do not exist yet
"set up tokens", "compile to CSS", "check contrast"humane:design-tokenshumane:layout-rules, which only flags the pair
"pick a font", "does this font have Cyrillic"humane:type-specimenhumane:design-tokens, until a family is chosen
"generate on-brand images"humane:brand-illustratehumane:brandkit, which explores rather than produces
"what should this button say", "rewrite this error"humane:ux-writing—
"polish this text", "fix typography", "add nonbreaking spaces"humane:typographyhumane:ux-writing if the wording itself is still open
"add meaningful feedback", "design this micro-interaction", "make this state change satisfying"humane:semantic-micro-interactionshumane:sound-effects, which owns audio assets rather than the interaction meaning
"turn these ideas into a scroll story", "build a scrollytelling sequence"humane:narrative-scrollytellinghumane:semantic-micro-interactions, which owns bounded component feedback rather than narrative order
"generate UI sounds", "make a feedback sound palette"humane:sound-effectshumane:semantic-micro-interactions when the state-change meaning is still open
"how does this land", "test this tagline"humane:respondent-panelhumane:persona-review — strangers, not experts
"review this PRD as an engineer / investor"humane:persona-reviewhumane:respondent-panel — experts, not strangers
"review this UI", "check this dashboard"humane:layout-rules—
"usability audit", "heuristic evaluation"humane:nielsen-heuristics—
"can someone actually complete this", "test this flow"humane:walkthroughhumane:nielsen-heuristics — attempt, not inspect
"what's the transformation"humane:before-after—
"review this properly", "full review", "holistic audit"humane:reviewany single domain skill

When more than one row matches, humane:review wins — it is the one skill allowed to run several domains and consolidate them into a single verdict. When the user names a skill explicitly, their words win over this table.

What the graph keys mean

Three keys, three different relationships. Do not collapse them.

  • handoffs: — a peer route across an ownership boundary, with the when: that opens it. humane:layout-rules hands a contrast pair to humane:design-tokens because it does not own the fix.
  • accepts: — the receiving skill's acknowledgement of that route. Every handoff has one; the repo's test suite fails the build if it does not.
  • orchestrates: — a call, not a handoff. humane:review runs its domain skills and consolidates their findings. No reciprocity is owed, and the domain skills are unaware of it.

Cycles are legal and deliberate. humane:ux-writing and humane:respondent-panel route both ways: writing hands copy out for reactions, reactions hand the rewrite back.

Honesty guards

  • A handoff is a condition, not an instruction. Each edge carries a when:. If the condition did not occur, the cycle ends here — do not run the next skill for completeness. A humane:layout-rules pass that found no contrast issue does not hand to humane:design-tokens.
  • Ask before continuing. A handoff names what could come next. The user decides whether it runs.
  • The graph is not coverage. Following every edge is not a full review, and saying "I ran the cycle" is only
阅读完整来源 ↗