Testing plan: a contribution has no outer loop, and its arming comment answers its first gate

Derived from the approved bugfix.md and design.md, before tasks.md — each task's _Test:_ names a row below. Authored at test-planning, completed at verification.

This file is executable content. It names commands an agent will run. No credential appears here, by value or by reference: every test runs offline against a fake GitHub integration and a temporary checkout.

Test matrix

#	Type	Applies?	Scope / what it proves	Where it runs
T1	Unit	yes	the checklist, the parse, the confirmation and the frozen record for a contribution — and unchanged for the outer loop	uv run --project cli python -m pytest -q cli/tests/test_graph_contribution.py
T2	Integration (scenario)	yes	a real GraphLink over a real Runtime and a fake GitHub: on_spawn carries an arming comment's goal to phase-selection; no goal parks; a respawn changes nothing	uv run --project cli python -m pytest -q cli/tests/test_graph_contribution.py -k spawn
T3	Integration (dispatcher)	yes	the spawning RoutedEvent reaches on_spawn, on the shared dispatcher both ingresses use	uv run --project cli python -m pytest -q cli/tests/test_graph_drive_integration.py
T4	Unit (prompt)	yes	$graph_context names no outer loop for a contribution, and still places one for every other loop; an agent start node is not evaluated at spawn	uv run --project cli python -m pytest -q cli/tests/test_graph_drive.py
T5	Contract (OpenAPI)	n/a — no API request or response shape changes; graph-state.json keeps its fields and surface keeps its type. test_api_contract_parity.py runs as part of T6 and would catch it if that were wrong.
T6	Regression (whole suite)	yes	the spawn path now runs an exit chain that never ran before; nothing else may move	make test
T7	UI / visual	n/a — no rendered UI; the two user-visible surfaces are a comment body and a prompt block, both asserted as text in T1/T4.
T8	Snapshot	n/a — no serialized artifact whose whole shape is asserted; the frozen record's changed field is asserted by name in T1.
T9	Performance / load	n/a — one extra exit-chain evaluation per fresh spawn of a human start node, on a path that has just started a tmux session and an LLM; unmeasurable beside it, and zero for every other spawn.
T10	Security / abuse case	yes	the three abuse cases in bugfix.md § Security considerations, each as a negative test	uv run --project cli python -m pytest -q cli/tests/test_graph_contribution.py -k unauthorized or surface
T11	Accessibility	n/a — no rendered UI.
T12	Migration / upgrade	yes	a work item frozen by 9.6.0 with surface: work-item still reads as before; a contribution written by this version carries "", which every reader already treats as the default	covered by T1 and by the pre-existing state round-trip tests in T6
T13	Manual exploratory	n/a — the reproduction needs a live daemon, a watched repository and credentials; T2 and T3 reproduce both halves deterministically and offline against the same code paths, which is stronger evidence than one manual run.
T14	Lint / typecheck / docs parity	yes	ruff, ruff-format, pyright, config validation, markdownlint and test_docs_parity.py	make check

Scenarios & requirement trace

Row	Requirement(s)	Scenario / case
T1	R1.1	the contribution checklist carries no outer-loop-on-pull-request row and says where the conversation happens
T1	R1.2	the work-item loop's checklist still carries it, worded as before
T1	R1.3, R1.4	a reply ticking the token on a contribution: pointer advances, state.surface == "", frozen surface == "", no phase declared away, confirmation silent about a surface
T2	R2.1	a spawn whose arming comment states the goal lands the item at phase-selection, checklist posted, binding recorded
T2	R2.2	a spawn with no goal parks at goal-definition carrying the gate's own reason
T2	R2.3	a second on_spawn re-records the binding and leaves every node record untouched, even once the goal has become answerable
T3	R2.1	the dispatcher hands the spawning RoutedEvent to the graph coupling
T4	R1.5	a contribution's prompt block says "a contribution has no outer loop" and never "the outer loop's artifacts"
T4	R1.2	a work item's prompt block still places the outer loop on its chosen surface
T4	R2.4	an agent start node is entered and not evaluated: not parked, no block recorded, one attempt
T10	abuse 1	an unauthorized goal in the thread leaves the gate waiting (pre-existing coverage, re-run)
T10	abuse 2	the injected surface token is inert (same case as T1/R1.3)
T10	abuse 3	an integration outage during the gate leaves the item waiting, never guessing (pre-existing coverage, re-run)
T6	R2.5	the whole suite: every other spawn path, every other loop, unchanged
T14	R3.1	docs parity over the changed CLI page and capability docs

Verification environment

A temporary directory per test (tmp_path) holding docs/specs/issue-9/, the shipped graphs from the package, and a fake GitHub integration injected at the_loop.graph.integrations.resolve. No network, no credentials, no tmux; the dispatcher rows use the suite's FakeTmux and StubInteractiveAdapter.

Verification results

Executed at verification — see evidence/.

#	Result	Evidence
T1	pass	evidence/unit.md
T2	pass	evidence/unit.md
T3	pass	evidence/integration.md
T4	pass	evidence/unit.md
T6	pass	evidence/check.md
T10	pass	evidence/unit.md
T12	pass	evidence/unit.md
T14	pass	evidence/check.md