Overview
Per clig (opens in new window):
"Send output to
stdout. The primary output for your command should go tostdout. Anything that is machine readable should also go tostdout.""Send messaging to
stderr. Log messages, errors, and so on should all be sent tostderr."
Tests can only assert on output that is stable, separated, and documented. This skill defines those conventions; sister skills (bats-testing, tui-snapshot-tester) test against them.
When to use
Convention 1 - Exit codes
Per clig (opens in new window):
"Return zero exit code on success, non-zero on failure."
"Map non-zero codes to important failure modes for script integration."
Exit code | Meaning
----------+----------------------------
0 | Success
1 | General error (catch-all)
2 | Misuse (bad flag, bad usage)
3 | Resource not found
4 | Permission denied
5 | Network / external failure
... documented per CLITest pattern:
@test "exit 0 on success" {
run mycli list
[ "$status" -eq 0 ]
}
@test "exit 2 on bad flag" {
run mycli --nonexistent
[ "$status" -eq 2 ]
}Don't test [ "$status" -ne 0 ] - assert the specific code. Otherwise refactors silently change the contract.
Convention 2 - stdout vs stderr
Per clig (opens in new window): "This separation ensures piped commands receive only data, not messages."
# Good: data on stdout, message on stderr
$ mycli list 2>/dev/null
alice
bob
$ mycli list >/dev/null
fetched 2 users in 0.3sTest pattern (bats):
@test "list emits names on stdout, status on stderr" {
run --separate-stderr mycli list
[ "$status" -eq 0 ]
[ "$output" = $'alice\nbob' ]
[[ "$stderr" == *"fetched 2 users"* ]]
}run --separate-stderr (Bats 1.5+) splits the streams; without it, $output mixes both.
Convention 3 - Machine-readable mode
Per clig (opens in new window):
"Display output as formatted JSON if
--jsonis passed."
"If human-readable output breaks machine-readable output, use
--plainto display output in plain, tabular text format for integration with tools likegreporawk."
"Encourage your users to use
--plainor--jsonin scripts to keep output stable."
$ mycli list --json | jq '.[] | .name'
"alice"
"bob"
$ mycli list --plain
alice active 2026-04-15
bob active 2026-04-20Test pattern: assert against --json, never against the human-formatted default. Human output is allowed to evolve; JSON is the contract.
@test "list --json contract" {
run mycli list --json
assert_success
echo "$output" | jq -e '.[0] | has("name") and has("status")'
}Convention 4 - Determinism
Stable output for tests requires:
# Bad: randomized order, locale-dependent date
$ mycli list
bob 2026/04/20
alice 2026/04/15
# Good: sorted, ISO 8601
$ mycli list --plain
alice 2026-04-15
bob 2026-04-20Test pattern: golden-file comparison with sorted output.
Conventions 5-8 - color/TTY, verbosity, --help/--version, documenting the contract
The remaining four conventions and their bats test patterns are in references/conventions-5-8.md.
Anti-patterns
| Anti-pattern | Why it fails | Fix |
|---|---|---|
Tests assert status -ne 0 | Refactors silently change error code; scripts break. | Assert specific code (Convention 1). |
| Errors on stdout | Pipes carry error text into downstream parsers. | stderr (Convention 2). |
| Tests assert against default human output | Human output evolves; tests churn. | Test against --json (Convention 3). |
| Timestamps in default output | Tests fail on every run. | Verbose-mode only (Convention 4). |
| ANSI codes leak when piped | wc -l, grep, etc. break. | TTY check + NO_COLOR (Convention 5). |
| Progress bars in CI | CI logs flooded; tests can't assert frame counts. | TTY check (Convention 5). |
| Help output requires network / config | --help should always work. | Help is local + static. |
--version mixed with marketing | Tooling can't parse. | Pure version string (Convention 7). |
Limitations
References
Continuation of the output contract defined in SKILL.md (color/TTY, verbosity, --help/--version, and documenting the contract). All cited to the Command Line Interface Guidelines (opens in new window).
Convention 5 - Color & TTY
Disable color when stdout/stderr is not a TTY, NO_COLOR is set (per no-color.org (opens in new window)), TERM=dumb, or --no-color is passed. Progress bars / spinners only on a TTY - they pollute CI logs and break wc -l assertions.
@test "no ANSI codes when piped" {
run bash -c 'mycli list | cat'
[ "$status" -eq 0 ]
refute_output --regexp $'\\x1b\\['
}
@test "NO_COLOR honored" {
NO_COLOR=1 run mycli list
refute_output --regexp $'\\x1b\\['
}Convention 6 - Quiet & verbose
A -q option suppresses all non-essential output; creator-only detail appears only in verbose mode.
Mode | Flag | Use
----------+--------------+--------------------
Quiet | -q / --quiet | Scripts; only data + errors
Default | (none) | Interactive; status messages on stderr
Verbose | -v | Debug info on stderr
Debug | -vv / --debug| Internal traces on stderr@test "-q suppresses status messages" {
run --separate-stderr mycli -q list
[ "$status" -eq 0 ]
[ -n "$output" ] # data still on stdout
[ -z "$stderr" ] # no status messages
}Convention 7 - --help and --version
-h/--help shows full help and works appended to any subcommand. --version must be machine-parseable (mycli 1.2.3), never Welcome to mycli! Version 1.2.3.
@test "--help exits 0 and mentions Usage" {
run mycli --help
[ "$status" -eq 0 ]
[[ "$output" == *"Usage:"* ]]
}
@test "--version emits machine-parseable version" {
run mycli --version
[ "$status" -eq 0 ]
[[ "$output" =~ ^[0-9]+\.[0-9]+\.[0-9]+ ]]
}Convention 8 - Document the contract
Each CLI repo keeps a CONVENTIONS.md (or README section). Tests assert against it; if the contract changes, the document and the tests update in the same PR.
## Output contract
- Exit codes: 0 ok, 2 bad usage, 3 not found, 4 perm denied, 5 network.
- `stdout` = primary data; `stderr` = status / errors / progress.
- `--json` is the stable machine contract; default human output may evolve.
- `NO_COLOR` and TTY detection respected.
- All dates in `--json` are ISO 8601 UTC.