🤖 This message was generated by kimi-k3. ## Problem `tests/integration/test_completion.py::test_completion_output` compares a generated bash completion script byte-for-byte against the checked-in fixture `tests/integration/test_completion/expected_script.sh`. The shared Starcraft build tooling (canonical/starbase `common.mk`) formats every git-tracked file that `file --mime-type` reports as a shell script using `shfmt -w`, explicitly excluding `*.sh.j2` templates so the Jinja2 source (`craft_cli/completion/bash_completion.sh.j2`) is not corrupted. However, the *expected output* fixture is still a plain shell script, so shfmt rewrites it (e.g. `_complete_testcraft(){` → `_complete_testcraft() {`, `all_cmds=( ls cp )` → `all_cmds=(ls cp)`), while the generated output continues to follow the template's original formatting. The byte-for-byte comparison then fails. ## Fix - Format the fixture with `shfmt` so it matches what repo tooling produces. - Make the test normalize the generated script through `shfmt -i 2` before comparing, so the comparison is formatting-insensitive on both sides. The test skips if `shfmt` is not installed. This keeps the test passing regardless of whether the fixture has been reformatted by repo tooling. ## Validation - `uv run pytest tests/integration/test_completion.py` — 2 passed - `uv run pytest tests/integration -q` — 188 passed - `uv run pytest tests/unit -q` — 684 passed, 1 skipped ## Provenance Discovered while validating the Starbase sync merge for this repository: the updated `common.mk` from canonical/starbase adds `format-shfmt`/`lint-shfmt` targets covering all tracked shell scripts. This is a preparation PR for that starbase update.