Verification: multi-modal messages are forwarded to the session, never parsed by the loop

The verification node's captured output. The results table lives in testing-plan.md § Verification results, against the matrix rows it planned; this file is the raw evidence those rows point at.
Every command was run from the repository root. uv is invoked as python3 -m uv (and make with /usr/local/bin first on PATH) because the container's own uv (0.8.17) predates this repo's required-version pin (>=0.12,<0.13); pip install "uv>=0.12,<0.13" supplied 0.12.17. No behaviour of the repo's tooling changed. No network call was made by any test: every fetch goes through an injected fake, and the credential variables the tests set are the literals in the test source.

T1 / T8 — the unit module, red then green

Red — the test module was written before the module it asserts:

text

$ python3 -m uv run --project cli python -m pytest -q cli/tests/test_attachments.py
    from the_loop.channels import attachments as mod
E   ImportError: cannot import name 'attachments' from 'the_loop.channels'
ERROR cli/tests/test_attachments.py
1 error in 0.18s

Green, after cli/the_loop/channels/attachments.py landed:

text

$ python3 -m uv run --project cli python -m pytest -q cli/tests/test_attachments.py
.......................                                                  [100%]
23 passed in 0.06s

The abuse-case selection (T8), across both new modules:

text

$ python3 -m uv run --project cli python -m pytest -q cli/tests/test_attachments.py \
    cli/tests/test_attachments_integration.py \
    -k "off_host or redirect or cap or hostile or untrusted or stranger or permalink"
13 passed, 26 deselected in 0.12s

T2 — the integration scenarios, red then green

Red — the scenarios were written before the pipeline, the dispatcher and the probe learned about files:

text

$ python3 -m uv run --project cli python -m pytest -q cli/tests/test_attachments_integration.py
E   ImportError: cannot import name 'attachment_findings' from 'the_loop.channels.slack'
ERROR cli/tests/test_attachments_integration.py
1 error in 0.19s

Green, after the wiring; the sixteen scenarios, by their Gherkin titles:

text

$ python3 -m uv run --project cli python -m pytest -q cli/tests/test_attachments.py \
    cli/tests/test_attachments_integration.py
.......................................                                  [100%]
39 passed in 2.27s

Scenario	Row
a Slack image reaches the session as a path and the ticket as a link	T2
a voice note reaches the session and the ticket as Slack's transcript	T2
a transcript Slack is still producing is re-read before delivery	T2
a message that is only an image is delivered	T2
a file that cannot be fetched is still named, and the message is delivered	T2
a stranger's file is never fetched	T8
a poll-read reply carries its files	T2
a gate answer with a file names it on the record	T2
a recorded thread names the files its messages carried	T2
a kickoff with a file links it from the issue body	T2
a GitHub comment's screenshot reaches the pane as a path after the excerpt	T2
a private asset with no token is named, not fetched	T2
an asset seen again is reused from disk	T2
a prompt whose event carries no attachment is byte-identical to before	T10
the manifest declares files:read	T2
a missing files:read is one finding; unreadable scopes are none	T2

Two scenarios were corrected between red and green for the test's shape, not the code's: the kickoff's issue body ends with the ledger's own attribution footer, so the 📎 line is asserted present after the message rather than last; and the pipeline's drop record for a stranger is {"outcome": "unauthorized-actor"}, not a dropped wrapper.

T10 / T12 — the touched suites, parity and the event catalog

text

$ python3 -m uv run --project cli python -m pytest -q cli/tests/test_state_portability.py \
    cli/tests/test_channels_mentions.py cli/tests/test_channels_dm.py \
    cli/tests/test_doctor_slack.py cli/tests/test_interaction_integration.py \
    cli/tests/test_channels.py cli/tests/test_docs_parity.py cli/tests/test_eventlog.py
231 passed in 2.96s

Four tests in that set went red first and were made honest rather than green: test_channels_dm.py (two probe tests), test_channels_mentions.py (the mention finding) and test_doctor_slack.py (the clean app) pin an exact findings count against a fake app whose granted scopes predate files:read; each fixture now grants the scope, so a "clean" app is one that can fetch a file, and the absence is pinned as issue-416's own finding in the integration module. test_eventlog.py caught the two new event types before they were catalogued (self-review F1).

text

$ python3 -m uv run python scripts/validate_config.py
VALID   .the-loop/harness-config.yaml
VALID   skills/the-loop/templates/harness-config.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

The full `make check`

text

$ PATH=/usr/local/bin:$PATH make check
uv run ruff check cli hooks
All checks passed!
npx --yes markdownlint-cli2@0.18.1 "**/*.md"
Summary: 0 error(s)
uv run ruff format --check cli hooks
350 files already formatted
uv run pyright cli
0 errors, 0 warnings, 0 informations
uv run python scripts/validate_config.py
VALID   … (all six, as above)
uv run --project cli python -m pytest -q cli
FAILED cli/tests/test_instance.py::test_an_unnamed_instance_spawns_with_the_argv_it_used_before
FAILED cli/tests/test_instance.py::test_a_spawn_for_a_work_item_exports_its_ref
FAILED cli/tests/test_instance.py::test_a_standing_session_carries_no_work_item
FAILED cli/tests/test_instance.py::test_only_the_configured_name_reaches_tmux_and_the_comment
4 failed, 4489 passed, 1 skipped in 196.81s (0:03:16)

The four failures are pre-existing and not this work item's. All four are in cli/tests/test_instance.py and concern the argv a tmux spawn is built with (the -e per-session environment, issue-410); this work item touches no spawn path. Reproduced on an untouched checkout of origin/main in a separate worktree:

text

$ git worktree add <scratch>/main-tree origin/main
$ cd <scratch>/main-tree && PYTHONPATH=cli python -m pytest -q cli/tests/test_instance.py
FAILED cli/tests/test_instance.py::test_an_unnamed_instance_spawns_with_the_argv_it_used_before
FAILED cli/tests/test_instance.py::test_a_spawn_for_a_work_item_exports_its_ref
FAILED cli/tests/test_instance.py::test_a_standing_session_carries_no_work_item
FAILED cli/tests/test_instance.py::test_only_the_configured_name_reaches_tmux_and_the_comment
4 failed, 59 passed in 0.26s

Identical set, none of this change applied — the same four issue-415 recorded, with the same visible cause (this container's tmux behaviour around -e). Recorded, not fixed: outside this work item's scope.

One side effect of the full suite worth naming: a test rewrites the timestamp inside the tracked fixture .the-loop/portable/github-octo-repo-15.json. It was restored with git checkout after each run and is not part of this change.

T11 — the operator's first-use check (not executed here)

No Slack workspace, app or token exists in this environment, and a private-repository asset needs a token with access; neither can be faked into meaning anything. The check is the operator's, on first use after re-installing the app with files:read:

In a work item's bound thread, mention the-loop with a screenshot attached. Expect: 👀 then ✅ on the message; a 📎 <name> (image/png, …) — <permalink> line on the ticket's record; the session's pane ends with Attachments (1) naming a path under <state.root>/local/attachments/<slug>/, and the session reads it.
Record a voice clip in the same thread and mention the-loop. Expect: the same 📎 line plus > Slack's transcript: … on the ticket; the pane's line quotes the same transcript. If Slack was still transcribing, the pane says so and names the saved clip.
On a private repository, comment on the work item with an image. Expect: the session's next prompt ends with an Attachments (1) section naming a saved path — or, with no GH_TOKEN/GITHUB_TOKEN in the daemon's environment, the URL with the transfer failed and the offer to try it.
the-loop channels status --probe on an app that was not re-installed. Expect: the files:read finding, and step 1's file named with the transfer failed rather than a path.