End-to-end plan

Every provider in this catalog is tested in the same tiers. A tier that cannot run is a skip with a reason, not a pass.

Tier	What it proves	Command	Needs
T0	This provider's unit and mock tests	`node tests/run.mjs` in the provider directory	Node >= 20
T1	Install, consent, pin, tamper, doctor, uninstall. No sandbox is created	`cli/e2e/providers/run.sh offline` and `python-sdk/e2e/opensandbox/provider_install_e2e.py`	git, Node, the SDK
T2	Create, exec, files, and kill through `abx`, with a provider that needs no cloud	`cli/e2e/providers/run.sh local-process`	Python 3
T3	The install line users type: `abx provider install alibaba-agent-sandbox/awesome-providers//<name>`	`cli/e2e/providers/run.sh catalog`	git and GitHub
T4	A real backend this plugin talks to. For OpenSandbox that is a local server	`local-up.sh up`, then `run.sh opensandbox` and `provider_e2e.py`	Docker for OpenSandbox
T5	The built-in Alibaba Cloud provider. Not a test of this catalog	`npm run e2e` in the CLI repo, `make e2e` in the SDK repo	cloud credentials

Rules

Throwaway HOME. Do not touch the developer's real ~/.abx.
Do not print API keys. provider config list must not contain them.
A feature the plugin does not declare must fail as E8004.
Do not use abx create "a sentence" as a provider test. Use abx --provider <id> create --template <image-or-name>.
Kill only the sandboxes this run created.

What a pull request includes

T0 passes. The README says what works, what is partial, and what returns E8004. T1 installs the new directory. T4 is required only when the plugin calls a backend you can start in the pull request.

The copy of this table that the CLI and SDK repositories keep is cli/e2e/providers/PLAN.md.