test: cover help rendering and the output layer's suppression rules

[patramsey]

	Before	After
cmd/help.go	0.0%	68.8%
internal/output/output.go	64.2%	70.3%
Repo total	69.2%	71.4%

On my earlier objection

I argued against testing help output, on the grounds that it becomes a change-detector — failing whenever someone rewords a string, never when anything breaks. That objection is about how the tests are written, not whether the code should be tested, so these assert structure rather than wording:

• Available commands appear; hidden ones do not — a hidden command in help advertises something unsupported
• Hidden flags aren't rendered; names, shorthands and usage strings survive
• printFilteredFlags shows only what the allow list permits
• With colour off, no escape sequence reaches the writer — help is routinely piped into a file, a pager, or an agent
• Long falls back to Short; a leaf command with no subcommands and no flags doesn't panic

Reword any description in this repo and these still pass. Delete a command from the list and they fail.

essentialGlobalFlagNames is the one place a literal name is pinned, deliberately: --dry-run and --yes gate destructive actions, so dropping either from subcommand help hides the safety controls on exactly the pages where someone is about to mutate something.

The output layer is all about suppression

Nearly every one of these guards something that must not appear:

• Hint is silent outside table mode — emitted into JSON it corrupts the document a caller is about to parse
• WarnBox degrades to plain prefixed lines rather than drawing a box into a pipe, but must not go silent; its warnings are the ones that earned extra weight
• Spinners are inert on a non-TTY, and Stop/Update stay safe to call. Stopping twice doesn't panic
• YAMLList omits nextPage when there is no next page, so callers test presence rather than comparing to zero
• DefaultConfig picks JSON when stdout isn't a terminal — the contract every piped invocation depends on

One test was worthless, and mutation testing caught it

The NO_COLOR presence cases passed vacuously. Under go test stdout isn't a TTY, so ColorEnabled() returns false whatever NO_COLOR does — a boolean misreading of the spec passed just as happily as the correct presence check.

They now set CLICOLOR_FORCE=1 alongside, which makes the two readings diverge: correct code returns false because NO_COLOR is checked first, a boolean reading returns true. The assertion can now fail, and it pins that NO_COLOR outranks CLICOLOR_FORCE into the bargain.

Verification

Eleven mutations, all caught:

Mutation		Mutation
NO_COLOR read as boolean	✅	Hint leaks into structured output	✅
env var precedence swapped	✅	WarnBox goes silent	✅
CLICOLOR_FORCE ignored	✅	hidden commands leak into help	✅
DefaultConfig always table	✅	hidden flags leak into help	✅
YAMLList emits zero nextPage	✅	--dry-run dropped from help	✅
(+ the NO_COLOR fix above)	✅

Also fixed a fixture bug the tests surfaced immediately: the first helpFixture built subcommands without a Run, and cobra's IsAvailableCommand reports false for those — so they were excluded from help for the wrong reason and the assertions would have passed vacuously.

• golangci-lint run — no issues
• go test -race -count=1 ./... — full suite clean
• 30 subtests confirmed running via -v

[codecov-commenter]

⚠️ Please install the to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!