.shellcheckrc

# ShellCheck configuration for this repository.
#
# Read by `shellcheck` for any file under this directory, so the CI job and a local
# run see the same rules. Three codes are disabled, and the bar for being here is
# narrow on purpose: EVERY instance in the tree must be an intentional pattern this
# repository already documents, verified one at a time. A code with even one genuine
# finding stays enabled and is fixed in the script instead. The counts below were
# taken with shellcheck 0.11.0 over `hooks/*.sh bin/* statusline/*.sh` on 2026-09-02;
# re-derive with `shellcheck -f gcc hooks/*.sh bin/* statusline/*.sh | grep -oE 'SC[0-9]+' | sort | uniq -c`.

# why: single-quoted strings that are meant to stay literal. All 25 instances are one
# of three deliberate shapes: a jq program whose `$name` is a jq variable bound by
# `--arg` (never a shell variable), a `sed` program, or a `printf` format string
# carrying literal backticks and `$` for the Markdown a hook writes. Double-quoting any
# of them would break it, which is the opposite of what this check suggests.
disable=SC2016

# why: positional field names in `IFS="$SEP" read -r a b c d` over a delimited record.
# All 16 instances name every field of the record even where the branch consumes only
# some, which is how the record's shape stays readable at the read site. They are not
# unused variables in the sense the check means; dropping them would make the field
# order implicit and the next reader would have to count separators.
disable=SC2034

# why: functions reached only through `trap`. All 7 instances are `cleanup` (six
# scripts, each with `trap cleanup EXIT`) or `flush_stage1`, which `cleanup` calls.
# ShellCheck cannot see a trap as an invocation, and the check says so itself:
# "or ignored if invoked indirectly".
disable=SC2329