Evidence — unit, abuse cases and the no-migration claim (issue-203)

Rows T1, T8 and T10 of ../testing-plan.md. Nothing here reaches the network: every URL is a hooks.slack.example fake, and the environment variable is set and deleted by name through monkeypatch. No redaction was needed — no real credential exists in this repository.

T1 — precedence, both transports

Resolution is driven through resolve("slack", config), not by constructing a provider: the config key is what an operator writes, and the wiring is the half that was missing.

$ uv run --project cli python -m pytest -q cli/tests/test_graph_integrations.py
.......................                                                  [100%]
23 passed in 0.11s

The four precedence cases (inline only · inline over env · env only · neither) run parametrised over webhook and sdk, so a future change cannot make the transports disagree about where a notification goes.

Red before green

The same command, run before any source file was touched:

FAILED cli/tests/test_graph_integrations.py::test_an_inline_url_is_used[webhook]
FAILED cli/tests/test_graph_integrations.py::test_an_inline_url_is_used[sdk]
FAILED cli/tests/test_graph_integrations.py::test_an_inline_url_wins_over_the_environment[webhook]
FAILED cli/tests/test_graph_integrations.py::test_an_inline_url_wins_over_the_environment[sdk]
FAILED cli/tests/test_graph_integrations.py::test_the_failure_names_both_remedies_and_not_the_url
FAILED cli/tests/test_graph_integrations.py::test_a_non_string_url_is_refused_by_the_schema
6 failed, 17 passed in 0.80s

test_an_empty_inline_url_falls_back_to_the_environment passed in both runs by construction — before the change the key was ignored entirely. It is kept as the guard that the fallback survives, and it earned its place: the first cut used str(section.get("url", "")), which lets a blank url: win over a working env var.

T8 — the three abuse cases

$ uv run --project cli python -m pytest -q cli/tests/test_graph_integrations.py \
    -k "refused or empty or remedies"
.........                                                                [100%]
9 passed, 14 deselected in 0.07s

Abuse case	Test
1 — a non-string url	test_a_non_string_url_is_refused_by_the_schema — validates the real .the-loop/cli-config.schema.json, and asserts additionalProperties: false still holds so the surface widened by exactly one named key
2 — a blank url	test_an_empty_inline_url_falls_back_to_the_environment
3 — the URL leaking	test_the_failure_names_both_remedies_and_not_the_url — asserts both sources appear in the message and that hooks.slack does not

The selector also picks up the pre-existing *_is_refused transport tests, which is why 9 rather than 3 run; all pass.

T10 — a pre-change config still validates, and nothing needs migrating

$ uv run python scripts/validate_config.py
VALID   .the-loop/harness-config.yaml
VALID   skills/the-loop/templates/harness-config.yaml
VALID   cli/the_loop/harness-config.default.yaml
VALID   .the-loop/collaborators.yaml
VALID   skills/the-loop/templates/collaborators.yaml
VALID   .the-loop/cli-config.yaml
VALID   skills/the-loop/templates/cli-config.yaml

Both CLI configs validate without an inline url — the property is optional, so a config written before this change is still a valid config.

migrations.CURRENT_CONFIG_VERSION is 0.4.0 before and after, and the diff of .the-loop/cli-config.schema.json touches no version field:

$ git diff .the-loop/cli-config.schema.json | grep -c version
0

That is the whole no-migration claim: an optional additive property is a superset of the old shape, and the version gate exists to refuse a shape the CLI no longer understands.

Documentation parity (issue-117's P1–P5), which is what forces the new schema leaf to be documented with a Type and a Default:

$ uv run --project cli python -m pytest -q cli/tests/test_docs_parity.py
.....                                                                    [100%]
5 passed in 0.05s