Hi @Mohamed-Elwasila, thanks for picking up this TODO 👍

I checked out this branch locally on Windows (Python 3.12, fresh venv from pyproject.toml) and ran the full file. All 16 tests pass, including the 3 new fish ones:

tests/test_completion/test_completion_option_colon.py ... 16 passed in 26.64s

A few observations from the review:

Things that look right

• The three tests cleanly mirror the existing *_zsh_* ones, which is exactly the right pattern — same fixture (colon_example.py), same _TYPER_COMPLETE_ARGS semantics (trailing-space ⇒ incomplete == "", no trailing-space ⇒ last token is the incomplete).
• The _TYPER_COMPLETE_FISH_ACTION="get-args" value matches what FishComplete.complete() checks for in typer/_completion_classes.py, so the tests are exercising the same code path that fish actually triggers.
• Removing the # TODO: tests for complete_fish comment is the right move — the TODO is now resolved by the new tests.

One suggestion that would strengthen the tests (optional — not blocking)

The current assert "alpine:hello" in result.stdout style verifies that the value appears, but it doesn't verify the fish-specific output format. FishComplete.format_completion produces value\thelp when help is present, and bare value when it isn't. So for the _all case you could additionally assert that the help text appears for the items in colon_example.image_desc that have descriptions:

assert "latest alpine image" in result.stdout          # help for alpine:latest
assert "fake image: for testing" in result.stdout       # help for alpine:hello

That would catch a regression in format_completion that the current value-only assertions would miss. Same idea applies to the zsh tests in this file, fwiw — they have the same gap — so it might make sense as a separate small follow-up rather than expanding this PR's scope.

Sorry the PR has been sitting for a while — bumping it for visibility.