warn on same-precedence prefix-pair traps (*/a b c, +-, etc.)

[danieljohnmorris]

Summary

*/a b c parses as (a/b)*c, not (a*b)/c. Same trap for /*, +-, -+. This is by design (outer prefix op binds inner subexpression as its left operand, matching the canonical +*a b c == (a*b)+c rule) but it consistently misleads agents reading the tokens left-to-right. Three persona logs in the assessment doc have hit it across three rerun sessions (logs-forensics entries at lines 844, 1056, 1651, 2833).

The token cost is real: a persona spends a turn debugging "why does my percentage formula give the wrong number?" then a second turn rewriting to r=*imp gl;pct=*r 100. A one-line hint at the terminal lets attempt #1 land correctly.

Manifesto framing: this is pure additive token-saving. The hint fires after a successful run, costs zero tokens when not relevant (suppressible with -nh), and saves a multi-turn debug loop when it is.

Repro before / after

$ ilo 'test a:n b:n c:n>n;*/a b c' test 6 2 3
9

Before: result is 9 (= (6/2)*3) with no warning. Persona reading the source as "multiply, then divide" expected 4 and silently believed the program.

After: same 9 result, plus

hint: `*/a b c` parses as `(a/b)*c` (inner prefix op binds first). For the other order, swap the ops (`/*a b c`) or bind: `r=*a b;/r c`

What's in the diff (per commit)

1. warn on same-precedence prefix-pair traps in collect_hints — adds the detector in src/main.rs. Walks the token stream from lexer::lex(source), fires on the four mixed same-precedence pairs (*,/), (/,*), (+,-), (-,+) when both ops are in prefix position (predecessor is not value-yielding). Same-op repeats and different-precedence pairs are skipped because they match the left-to-right reading. 12 unit tests pin the firing, non-firing, predecessor, and negative-literal cases.
2. add examples/prefix-mul-div.ilo for engine-harness coverage — three named functions exercise the trap, the swap, and the bind rewrite, with -- run: / -- out: assertions for tree / VM / Cranelift. Also serves as in-context learning material for agents who hit the trap and read examples to understand the fix.
3. document the same-precedence prefix-pair trap — SPEC.md gets a section after the canonical prefix-nesting examples, ai.txt gets a matching row in the compact AI spec, and skills/ilo/SKILL.md gets a one-liner in the core-syntax block. Matching site update at ilo-lang/site@57bf147.

Test plan

• cargo test --release --features cranelift — 22 collect_hints tests (12 new) pass, examples_engines runs the new example across tree / VM / Cranelift and passes.
• cargo fmt --all -- --check clean.
• cargo clippy --workspace --all-targets --features cranelift -- -D warnings clean.
• Manual smoke: ilo run examples/prefix-mul-div.ilo mul-div-trap 6 2 3 prints 9 + hint. ilo 'f a:n b:n c:n>n;+*a b c' f 2 3 4 prints 10 with NO hint (different precedence, intuition matches). ilo 'f a:n b:n c:n>n;++a b c' f 2 3 4 prints 9 with NO hint (same-op repeat).
• -nh suppresses the hint as expected.

Follow-ups

None. The fix is self-contained and the persona pattern is fully covered by the four shapes in the detector.

[codecov]

Codecov Report

❌ Patch coverage is 99.09910% with 1 line in your changes missing coverage. Please review.
✅ All tests successful. No failed tests found.

Files with missing lines	Patch %	Lines
src/main.rs	99.09%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!