docs: document lowercase-only identifier rule in SPEC and SKILL

[danieljohnmorris]

Summary

PR #262 documented the post-dot field-access relaxations (camelCase, leading-uppercase, snake_case, kebab-case) so JSON keys from real APIs just work after . and .?. What it didn't document, because it was implicit, is the baseline rule the relaxations are an exception to: identifiers at binding and call sites are lowercase ASCII only, optionally with hyphenated segments.

Anyone reading the dot-access section in SPEC.md would reasonably wonder what the baseline naming rule is. This PR makes it findable on its own anchor.

Repro before/after

Before this PR, the only mention of the lowercase-only rule was an indirect note at the bottom of the dot-access section: "top-level identifiers still follow the standard naming rules". The "standard naming rules" themselves were only the 1-3 char shorten-everything heuristic, with nothing about ASCII-lowercase, hyphens, or the post-dot exception.

After this PR, a new Identifier syntax subsection under Naming gives the formal grammar ([a-z][a-z0-9]*(-[a-z0-9]+)*), a worked OK/ERROR table covering capital letters, leading capitals, underscores, and leading hyphens, the ILO-L003 surface in the CLI with the exact suggestion shape, and an explicit cross-reference to the Field names at dot-access section.

Verified against the lexer:

$ echo 'runD = 1' | ilo run -
{"code":"ILO-L003", ... "suggestion":"identifiers must be lowercase ASCII; got 'runD' (capital 'D' at offset 3). Use lowercase, e.g. `run-d` or `rund`"}

The SPEC claim matches the implementation exactly.

What's in the diff

Commit 1: SPEC.md - new Identifier syntax subsection under Naming with formal grammar, OK/ERROR table, ILO-L003 mention, and link to Field names at dot-access.

Commit 2: SKILL.md + ai.txt - mirrors the rule into SKILL.md (compressed: agents see this first when ilo loads as a Claude plugin), plus the build.rs regeneration of ai.txt.

Test plan

• cargo build --release --features cranelift - build.rs regenerates ai.txt cleanly
• Verified runD produces ILO-L003 with the exact suggestion the SPEC documents
• Docs-only change, no test impact

Follow-ups

None.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ All tests successful. No failed tests found.

📢 Thoughts on this report? Let us know!