# Manual checks

These tools are opt-in. The full test gate excludes them because they require
credentials, a real SSH host, an installed package, or benchmark artifacts.
Run commands from the repository root inside `nix develop`.

| Tool | Invocation and prerequisites | Result |
| --- | --- | --- |
| TUI screenshots | `ALBEDO_SHOT_DIR=/tmp/shots go -C cli test ./internal/tui -run TestShot`, then `python3 test/manual/tui_shot.py /tmp/shots` | Needs `rsvg-convert`. Draws each `TestShot*` screen (what the real `View()` emits, in a catppuccin terminal's colors) as a PNG beside its `.ans`, to look at before and after a layout change. Add a screen by calling `shot(t, name, view)` from a test after `shotTerminal(t)`. |
| Packaged binary | `python3 test/manual/nix_package_smoke.py /absolute/path/to/bin/albedo` | Checks the installed client and daemon outside the checkout with a restricted PATH. Fails on an invalid package. |
| Remote kernel | `python3 test/manual/remote_ssh.py user@host` | Requires key authentication and remote Python 3.11 or newer. Exercises staging, kernel boot, tool calls, and connection reuse. |
| Kernel memory | `python3 test/manual/kernel_memory_benchmark.py --scale 1 10 --output /tmp/kernel-memory.json` | Measures boot and workload memory. Writes measurements for comparison, rather than imposing a universal memory threshold. |
| Active output | `python3 test/manual/active_output_benchmark.py --baseline /tmp/baseline/albedo-daemon --candidate /tmp/candidate/albedo-daemon --output /tmp/active-output-results.json` | Requires two exported daemon builds and Linux `/proc` for memory measurements. Compares attachment and hydration performance using a gated local provider. |
| Interval implementation | `python3 test/manual/verify_intervals.py /directory/containing/interval_set.py` | Runs five unittest scenarios, including randomized interval algebra. Exits unsuccessfully when the candidate violates the contract. |
| Provider benchmark | `gleam run -m manual/benchmark` | Requires `ALBEDO_BENCH_URL`, `ALBEDO_BENCH_MODEL`, and `ALBEDO_BENCH_KEY`. Optional `ALBEDO_BENCH_PROTOCOL` selects a protocol. Prints request timings and memory measurements. |
| Stream benchmark | `gleam run -m manual/stream_benchmark` | Offline. Frames and reduces synthetic Responses and Chat Completions streams and encodes a 1 MB request, printing time per run and throughput. Compare before and after a change to `openai_api`. |
| Stream rate benchmark | `python3 test/manual/mock_openai_server.py` in one terminal, then `ERL_FLAGS="<vm_defaults from priv/bin/albedo-daemon>" gleam run -m manual/stream_rate_benchmark` | Local. Streams Responses, Chat Completions, Claude, Vertex, and Cloud Code Assist through the real transport from a mock server at 500 tokens/s and unthrottled, or `ALBEDO_BENCH_PARALLEL` streams at once, and prints VM CPU, reductions, and send-to-callback latency per token. `ALBEDO_BENCH_TOOLS` streams a share of the deltas as tool-call arguments. For https, start the mock with a certificate and key and set `ALBEDO_BENCH_ORIGIN=https://localhost:PORT` and `ALBEDO_BENCH_CA` to the issuing CA. |
| Copy probe | `gleam run -m manual/copy_probe` | Offline. Boots a runtime and one default session and prints what a spawn over the runtime handle, the installed list, the session, and each managed plugin copies, in KiB. Compare before and after moving a catalog between processes. |
| Live coding | `gleam run -m manual/coding` | Requires the benchmark URL, model, and key, plus `ALBEDO_LIVE_WORKSPACE` and `ALBEDO_LIVE_DATABASE`. Optional `ALBEDO_LIVE_PROTOCOL` selects a protocol. Runs a real model-driven coding scenario. |

`albedo_openai_bench.erl` supports the provider benchmark and live coding tool,
and `albedo_openai_rate_bench.erl` the stream rate benchmark.
`albedo_live_coding.erl` supports the live coding tool. They are native helpers,
not independent suites.

The `glinter-review.toml` and `glinter-all.toml` files configure advisory lint
profiles. Run `test/gleam-lint.sh --review` or `test/gleam-lint.sh --all`.
See [the lint documentation](../../robot-docs/gleam-lint.md) for interpretation.

Do not bulk-import these scripts as a smoke check. Some consume command-line
arguments or execute immediately on import. Ruff and ty cover their source in
the normal gate; exercise their documented commands when changing them.