MOB-5: ~MOB @foo raises a clear error when assigns is not in scope

[GenericJam]

Fixes MOB-5. Follow-up to the 0.7.11 LiveView-ergonomics release.

Problem

0.7.11 added @foo → assigns.foo sugar in ~MOB. But the idiomatic way to factor mob UI — including the guide's own example — is plain functions with positional args:

def card(title, children), do: ~MOB"<Text text={title} />"

There's no assigns in such a function. The moment a user follows the new "@assigns shorthand" docs and writes @title inside one, it expands to assigns.title → cryptic "undefined variable assigns". (Same trap if a render param is named socket instead of assigns.) A user hit exactly this: a wave of "no assigns" errors, then "unused assigns" warnings while trying to bridge the two styles.

What Phoenix does

Phoenix.Component.sigil_H guards with Macro.Env.has_var?(__CALLER__, {:assigns, nil}) and raises "~H requires a variable named assigns..." — a clear compile error instead of a raw undefined-variable. mob's sigil has __CALLER__ too, so it can do the same.

Fix

• Mob.Sigil: before rewriting @foo, check Macro.Env.has_var?(caller, {:assigns, nil}). If assigns isn't bound, raise a CompileError naming the fix — "@title requires a variable named assigns in scope … use {title} instead of @title." Applies to attrs, {expr} children, and :if/:for. Only @-using templates trigger it; a static ~MOB(<Text text="hi"/>) in a positional-arg helper still compiles (unlike ~H, which always needs assigns).
• Docs: scope @ to render(assigns), steer composites to positional {var}, and add the common-mistake note (including "name the render param assigns").

Tests

@foo with no assigns raises (asserts the message names {title}); :if={@flag} with no assigns raises; a static template compiles without assigns; a positional-arg helper ({title}) compiles + runs without assigns. Existing @assign tests (which bind assigns locally) stay green. Full suite: 964 pass; format + credo clean.

Not included

No version bump — release timing is the maintainer's call (would be 0.7.12).

🤖 Generated with Claude Code

[GenericJam]

Reviewed the diff, tests, and CI. Recommend merge.

Correctness: the guard sits inside the prewalk, so it only fires when a @name node is actually present — static templates and positional-arg helpers using {var} compile untouched. caller is threaded correctly from the sigil __CALLER__, and Macro.Env.has_var?(caller, {:assigns, nil}) is the same check Phoenix.Component.sigil_H uses, so it is proven.

Tests: targeted and real — @foo without assigns raises (asserting the message names {title}), :if={@flag} raises, a static template compiles, and a positional helper both compiles and runs. Guard + happy path both covered.

CI: fully green (Elixir 1.19/OTP 28, native formatters, mix_audit, GitGuardian); mergeable/CLEAN.

Minor, non-blocking: the guard also fires for a genuine module attribute @title used inside a sigil in a non-assigns function. That is pre-existing @-shadowing (same as HEEx) and the new error is still clearer than the old undefined-variable crash, so not worth holding the PR.

Coordination note (docs anchor): the new blockquote links to [function composites](#pure-elixir-composite-components). PR #57 renames that section heading to "Defining your own components", which removes that anchor. I am updating #57 to keep the #pure-elixir-composite-components anchor alive so this link does not dangle regardless of merge order. Suggest merging this (#56) first, then #57 rebases on top.

(Posting as a comment rather than an approval — GitHub blocks approving your own PR.)