qa.md

QA

• Running colab with no args prints help.
• Top-level help shows only session, run, fs, status, ai, auth, settings, and completions.
• continue, distribute, slurp, fleet, release, agent, doctor, config, tools, runtime, mount, exec, env, and bug-report stay out of production help.
• --color stays out of normal help; --no-color remains.
• status renders a human panel, not raw JSON.
• status --json, settings skills list --json, and ai tools list --json contain no ANSI.
• -v, -vv, and -vvv print debug1/debug2/debug3 lines to stderr.
• --json -v status keeps stdout valid JSON and writes debug lines to stderr.
• --quiet -vvv status suppresses debug output.
• Verbose output redacts bearer tokens, cookies, Colab tokens, and token-like query values.
• settings renders sectioned settings.
• settings, settings ui, and settings experiments allow multiple edits before s saves.
• Settings Enter opens/toggles, b/Esc goes back, q exits, and the footer lists only implemented keys.
• settings ui set and settings experiments set direct commands still persist.
• Multi-login cannot be enabled while Distribute is off.
• settings skills list renders an agent/tool catalog, not core commands.
• ai tools list renders the same agent-facing catalog without debug booleans.
• run ast, ai mcp, ai plan, and ai run fail with experimental feature disabled until enabled.
• distribute and continue fail with experimental feature disabled until enabled.
• secret stays hidden from default help and fails with experimental feature disabled until Secrets bridge is enabled.
• run ... --env KEY is gated by Secrets bridge and never prints secret values in normal, verbose, JSON, log, or support output covered by tests.
• ai tools list shows secret.inject and run.with_env only when Secrets bridge is enabled.
• run pip install, run pip freeze, and run pip restore parse.
• session kernel list/current/select/specs/start/interrupt/restart/shutdown/refresh parse.
• status kernel --all and status kernel --refresh parse.
• Kernel list output is aligned and does not print raw debug booleans.
• run pkg routes through active kernel language.
• run pip is blocked when cached active kernel metadata is Julia or R.
• run julia and run r parser paths remain hidden from normal help unless dynamic kernel-aware help intentionally exposes them.
• run --help uses cached kernel metadata and does not contact Colab.
• run repl uses local line editing and kernel execution; arrow keys do not reach the remote kernel as escape bytes.
• echo "print('hello')" | colab run repl executes once and exits.
• run shell connects through /colab/tty where supported and does not assume /api/terminals.
• echo "echo HELLO" | colab run shell sends exit\n and does not hang.
• Terminal raw mode is restored after shell and settings exits.
• Old run install, slurp, and fleet aliases stay hidden and print migration hints when used.
• NO_COLOR=1 colab status has no ANSI.
• CI=1 colab status check does not animate or prompt.
• session new keeps the existing prompt flow and renders a concise success card after assignment.
• Transient Colab assignment errors do not dump HTML unless --verbose is passed.
• fs drive mount shows staged progress in interactive terminals and maps kernel/approval failures to short errors.
• fs drive mount maps endpoint/network failures to a staged Drive error and hides raw reqwest text unless --verbose.
• auth login --method adc, auth login --method oauth2, status version, update, pay --dry-run, session logs, and session kernel restart --help parse.
• Non-TTY output stays plain.
• Shipyard remains a release tool first; no broad TUI is required for this pass.