build(lint): make gofmt a check, not a rewrite

[lancekrogers]

Summary

• Fixes CG-GATE-01: the lint recipe ran go fmt ./... (a silent rewrite), so the gate never actually failed on unformatted code and HEAD was not gofmt-clean.
• fmt now runs gofmt -l internal cmd and exits non-zero if anything is listed; fmt-fix preserves the old rewrite behavior for local use.
• Formatted the one offending file (internal/tui/grouped_cursor_test.go) so HEAD is clean.

Test plan

• gofmt -l internal cmd returns empty on this branch
• just lint exits 0 on the clean tree
• Verified just lint exits non-zero (fmt step fails before vet runs) when an unformatted file is present, then reverted the test injection
• just test unit — 182/182 tests passed

[obey-agent]

This narrows the fmt gate to cmd/ and internal/, but the old go fmt ./... it replaces covered the whole module — including the 16 committed .go files under tests/integration/. Unformatted test code will now pass this gate silently, which is the same "gate reports clean while tree is dirty" failure this PR is fixing, just moved to tests/. It matters because lint runs in CI via just lint (test.yaml, release.yaml). Change to gofmt -l . (gofmt only inspects .go files, so bin//dist//out//completions/ are ignored) or gofmt -l cmd internal tests. This also aligns the checker with fmt-fix's go fmt ./..., which already formats tests/.

[lancekrogers]

Fixed. Gate now uses gofmt -l . (matches fmt-fix's go fmt ./... scope, and gofmt only touches .go files so bin/dist/out/completions are naturally excluded). Formatted the 8 flagged files under tests/integration/.

Re-verified: gofmt -l . clean at HEAD (b3b2d97), just lint exits 0 on the clean tree, and confirmed the negative case directly — injected an unformatted function into tests/integration/helpers.go, just lint failed with exit 1 (listing the file, pointing to fmt-fix), then reverted and reconfirmed clean. just test unit: 182/182 passed. Pushed to fix/camp-graph-lint at b3b2d97.

[obey-agent]

Staff Review — Verdict: REQUEST CHANGES

Overview
The PR correctly diagnoses the gate bug: the old fmt recipe ran go fmt ./..., a silent rewrite that mutated files during lint and therefore never failed the gate on unformatted code. Splitting into a checking fmt (gofmt -l + non-zero exit) and a rewriting fmt-fix is the right shape, and the bash-shebang implementation is portable and defensive.

Key Findings

• (major) Coverage regression. The old go fmt ./... formatted every package in the module, including tests/integration/ (16 committed .go files). The new gofmt -l internal cmd only walks cmd/ and internal/, so unformatted code under tests/ passes the gate silently — the same failure mode this PR fixes, relocated to tests/. Since lint is wired into CI (test.yaml, release.yaml both run: just lint), the hole is load-bearing. Fix: gofmt -l . (gofmt only reads .go files) or gofmt -l cmd internal tests. This also removes the asymmetry where fmt-fix (go fmt ./...) formats tests/ but fmt never verifies it.

What's Done Well

• Correct use of gofmt -l + explicit [[ -n "$unformatted" ]] -> exit 1 (gofmt -l exits 0 even when it lists files, so this is the right way to fail the gate); empty-output correctly passes.
• #!/usr/bin/env bash + set -euo pipefail makes the recipe robust across shells and bash-3.2/macOS compatible.
• Preserving the rewrite path as fmt-fix keeps local fixing ergonomic; actionable "Run 'just fmt-fix'" message.
• Reformatting grouped_cursor_test.go so HEAD is clean, plus the documented negative test, is the right diligence.

Staff Standard
One real gap between "checks the whole module" and "checks two of three source dirs." It's a one-token change to close, and until it's closed the gate ships with a blind spot over committed test code. Everything else about the change is production-quality.

[obey-agent]

Staff Review — Verdict: COMMENT (merge-ready)

Overview
The follow-up commit resolves the prior review blocker. The formatting gate now checks the whole module with gofmt -l ., matching the scope of fmt-fix, and the previously missed integration test files are gofmt-clean.

Key Findings

• No remaining blocking findings. The earlier coverage gap over tests/integration/ is fixed.

What's Done Well

• fmt now actually gates formatting instead of rewriting files silently.
• fmt-fix preserves the local rewrite workflow while keeping CI/check behavior strict.
• The message printed on failure is actionable and points directly to just fmt-fix.

Verification

• gofmt -l . returned empty output.
• just lint
• just test unit (182/182 tests passed)

Staff Standard
Yes. I would be comfortable submitting this as my own work now that the formatter check covers all committed Go packages in the module.

[obey-agent]

fmt-fix uses go fmt ./..., whose scope is narrower than the fmt gate's gofmt -l .: go fmt ./... skips testdata/, vendor/, and ./_-prefixed dirs, while gofmt -l . inspects them. Harmless now (no .go files exist in those locations), but if a .go fixture lands under testdata/ unformatted, the gate would fail with no fix-recipe path. Consider gofmt -w . here for exact check/fix symmetry, or soften the commit message's "matches fmt-fix scope" claim. Non-blocking.

[lancekrogers]

Fixed in 476e368. fmt-fix now runs gofmt -w ., matching the fmt gate's gofmt -l . whole-module scope, so a future fixture under testdata//vendor//dot-prefixed dirs can't produce a red gate with no repair path. HEAD stays gofmt-clean (just fmt green).

[obey-agent]

Staff Re-Review (head b3b2d97) — Verdict: COMMENT — prior blocker FIXED

Overview
Delta since my prior review (9f7b612): the justfile gate widened from gofmt -l internal cmd to gofmt -l ., plus 8 tests/integration/*.go files reformatted (pure gofmt column-alignment, no behavioral change).

Prior finding — FIXED and verified end-to-end
The major coverage gap is closed: the gate is now gofmt -l . (whole module, recursive), so the tests/ tree is covered and its previously-unformatted files were reformatted. I reproduced the gate against the head tree: clean -> rc 0; injected an unformatted file -> rc 1 listing it; injected a parse-error file -> rc 2 (not swallowed). HEAD is gofmt-clean, so the newly-armed gate goes green rather than shipping self-red.

New finding (minor, inline)
fmt-fix uses go fmt ./..., whose scope is narrower than the fmt gate's gofmt -l . (go fmt ./... skips testdata/, vendor/, and ./_-prefixed dirs). Harmless today (no .go files in those locations), but a future testdata/*.go fixture could produce a red gate with no fmt-fix path. The commit message's "matches fmt-fix scope" claim is slightly off.

What's Done Well
Correct check-vs-rewrite separation; strict-mode shell with a clear failure message naming offending files and pointing at just fmt-fix; HEAD is actually formatted so the gate goes green.

Staff Standard
Gate reliability confirmed (non-zero on drift, zero on clean, non-zero on parse error). The one remaining item is a minor scope-symmetry nit. Merge-ready. Leaving the Approve for Lance.

[obey-agent]

Staff Re-Review - Verdict: APPROVE

No blocking findings on the updated head 476e368.

The prior formatter-scope nit is resolved: fmt gates with gofmt -l ., and fmt-fix now rewrites the same whole-module scope with gofmt -w .. HEAD is gofmt-clean, so the gate no longer has a blind spot over directories that go fmt ./... would skip.

Verification performed:

• git diff --check ddd235c66e61620ec5db4ab7cc108af39827955d...HEAD
• gofmt -l . returned empty output
• just lint
• just test unit (182/182 tests passed)