Exit Codes & Output Streams

The CLI follows one contract everywhere:

Exit codes

Code	Meaning
0	Success
1	Runtime failure — API error, network failure, missing credentials, partial batch failure
2	Usage error — invalid flag value, unparsable input file, missing required flag combination

Scripts can rely on 2 meaning "fix the invocation" and 1 meaning "fix the environment or retry".

stdout vs stderr

• stdout carries data only: pretty layouts, tables, and the raw payload for --output json|yaml|csv.
• stderr carries everything else: progress/status lines, rate-limit retry warnings, deprecation notices, and errors.

This means machine-readable output always pipes cleanly:

airs runtime profiles list --output json | jq '.[].profile_name'
airs runtime bulk-scan --profile demo --file prompts.txt --output-file results.csv
# progress lines appear on the terminal (stderr) without corrupting the CSV

API errors include the HTTP status when available, plus a reminder that --debug captures the full (secret-redacted) request/response traffic.

Confirmation prompts

Destructive commands (runtime profiles delete, runtime topics delete, runtime topics revert, redteam targets delete, runtime profiles cleanup) ask for interactive confirmation before deleting:

• Interactive terminal (TTY): a Y/N confirm prompt appears (default: No). Declining prints Aborted and exits 0 — a no-op, not a failure.
• --force: bypasses the prompt entirely (for profiles delete and topics delete, --force additionally performs the API force-delete and pairs with --updated-by).
• Non-interactive (no TTY, e.g. CI or piped): the command refuses without --force and exits 2 — a script must state destructive intent explicitly. Exception: profiles cleanup without --force stays a safe preview (exit 0) so agents and JSON consumers can inspect before deleting.