damage: the command line says what a damage takes, not only that it exists

[donislawdev]

The window drew the settings of a chosen damage from the declaration on the day
damage arrived. The command line named the damages and stopped there, so
learning that zero-head takes bytes, and that bytes runs from 4 to 4096,
meant typing a wrong value and reading the refusal.

The parity guard cannot see that gap: it counts recipe keys, formats and
presets, and a damage parameter is none of the three.

What is here

tfg damage                  every damage, and what each one does to the bytes
tfg damage zero-head        what one of them takes
tfg damage --json           the same, for a script

The list carries a sentence rather than only a name, which is the difference
from tfg formats: png and zip say what they are and zero-head does not,
so a list of names would repeat what the --damage flag help already says. In
--json the parameters ride under the key tfg preset show already uses, in
the shape tfg formats already uses, because a damage parameter is a
format.Property.

Three things it does not do quietly

• The smallest file follows the settings, so the column is measured with the
  defaults and says so. That column got it wrong once for formats: until
  2026-08-04 MINIMUM was the structural floor and pdf announced 3265 while
  runs took 3286.
• An unknown name ends with 2, not 4. damage.UnknownError is already
  classified that way tool-wide, because a recipe naming one is refused while
  the recipe is read. One mistake, one code, matters more than matching the
  neighbouring command.
• Two names at once are refused. Measured while writing this: tfg formats png svg describes png, ignores svg and ends with zero. formats is not
  changed here, because that is a visible behaviour change to a released
  command.

Guards

Four, each with a mutation, all caught and re-proved after the refactor below.
The last one holds commandsTakingHelp against the dispatch itself: adding this
command meant remembering two hand kept lists in two files, and nothing asked.
The rule is mechanical rather than a list of exceptions, so version, license
and help are not exempted - they simply do not hand on args[1:].

Flattening beat raising a ceiling again: the command pushed the count of
functions nesting three deep from 51 to 52, and lifting the list builder out put
it back at 51.

Verification

• full go test ./... green, gofmt, go vet, staticcheck@v0.8.1 clean
• tools/journey.py 300 of 300
• tools/mutate/staleness.py: 940 patterns, each occurring exactly once
• argument order, unknown flag, empty argument and wrong case exercised on the
  built binary

🤖 Generated with Claude Code