Evidence — verification

Every applicable row of testing-plan.md, run on claude/github-issue-360-uav3nz with the fix in place.

T1, T2, T3, T10 — the argv, the critic seam, the verdict cache, the boundaries

$ uv run --project cli python -m pytest -q \
    cli/tests/test_critics.py cli/tests/test_modelchoice.py cli/tests/test_modelprobe.py
99 passed in 1.49s

$ uv run --project cli python -m pytest -q cli/tests/test_critics.py -k builtin_harness
1 passed, 51 deselected in 0.03s

The argv itself, after the fix — the same one-liner bugfix.md §Steps to reproduce uses:

$ uv run --project cli python -c "from the_loop.harness import CursorAgentAdapter as C; \
print(C().oneshot_argv('review this', model='gpt-5.6-sol'))"
['-p', 'review this', '--output-format', 'json', '--model', 'gpt-5.6-sol']

T13 — every gate CI runs

$ make check
uv run ruff check cli hooks
All checks passed!
npx --yes markdownlint-cli2@0.18.1 "**/*.md"
markdownlint-cli2 v0.18.1 (markdownlint v0.38.0)
Linting: 1103 file(s)
Summary: 0 error(s)
uv run ruff format --check cli hooks
301 files already formatted
uv run pyright cli
0 errors, 0 warnings, 0 informations
uv run python scripts/validate_config.py
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
uv run --project cli python -m pytest -q cli
3605 passed, 1 skipped in 159.34s (0:02:39)

Before this work item the suite was 3603 passed, 1 skipped; the two added tests are T1's argv assertion and T3's verdict-cache assertion. No test was skipped, disabled or deleted: three existing assertions were corrected from -m to --model, listed in tasks.md tasks 1 and 3.

T5 — the live cursor-agent run (external)

Not reproducible here: this container has no cursor-agent and no Cursor account, which is why T4/T5 are n/a in the matrix. The end-to-end proof is the reporter's, on cursor-agent 2026.09.10-fd3934a, quoted verbatim in issue-360:

$ cursor-agent -p "say OK" --output-format json -m auto --force
error: unknown option '-m'

$ cursor-agent -p "reply with exactly: OK" --output-format json --model auto --force
{"type":"result","subtype":"success","is_error":false,"result":"OK", ...}

The second line is the argv this change now produces, with the model in the same position.

Acceptance criteria

Criterion	Met by
R1.1 — a resolved model is passed as --model <name>	T1: test_cursor_oneshot_argv_uses_the_long_model_flag, test_a_model_resolves_through_the_adapters_flag
R1.2 — the full critic argv for harness: cursor + model:	T2: test_builtin_harness_derives_argv_from_the_adapter
R1.3 — no model, no flag: the argv is unchanged	T1: test_oneshot_argv_without_a_model_is_the_plain_one_shot_run, untouched and still passing
R1.4 — a regression test that fails before and passes after	red.md → this file
R2.1 — the model_flag rule in the capability doc	docs/capabilities/review-loop.md §Current behaviour, lint-clean at T13
R2.2 — a History row tracing it to issue-360	docs/capabilities/review-loop.md §History, first row