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 says	Start with	Not with
"something's missing / nothing works / fresh machine"	humane:setup	anything else — it diagnoses first
"what should I build", "what are people hiring this for"	humane:jtbd	a review skill; there is nothing to review yet
"mock this up", "show me the flow", "wireframe this"	humane:prototype	humane:design-tokens — the structure is not settled yet
"implement this with Astryx", "fit this to our design system"	humane:design-frameworks	humane:prototype — use the adapter after the product decision is settled enough to fit
"we need a brand / an identity"	humane:brandkit	humane:design-tokens — the tokens do not exist yet
"set up tokens", "compile to CSS", "check contrast"	humane:design-tokens	humane:layout-rules, which only flags the pair
"pick a font", "does this font have Cyrillic"	humane:type-specimen	humane:design-tokens, until a family is chosen
"generate on-brand images"	humane:brand-illustrate	humane:brandkit, which explores rather than produces
"what should this button say", "rewrite this error"	humane:ux-writing	—
"how does this land", "test this tagline"	humane:respondent-panel	humane:persona-review — strangers, not experts
"review this PRD as an engineer / investor"	humane:persona-review	humane: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:walkthrough	humane:nielsen-heuristics — attempt, not inspect
"what's the transformation"	humane:before-after	—
"review this properly", "full review", "holistic audit"	humane:review	any 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 true for the skills that actually ran. Name the ones that did not.
• Never invent an edge. If scripts/graph.py shows no route, there is no route. Add one to the frontmatter — on both sides — rather than improvising it in conversation.

Claude Code extras: the skills are invoked as humane:<skill>. Announce each one as it starts, so a multi-skill cycle narrates itself.