Testing plan: multi-repo work items — the outer loop stays in the origin repo

Derived from requirements.md and design.md, reviewed with the design at one gate, and locked before tasks.md. Executed at the verification node, which fills in § Verification results below.

Test matrix

#	Type	Applies?	Scope / what it proves	Where it runs
T1	Unit	yes	repo_state_key, inner_loop_state_dir, await_inner_loops (declared repos, both layouts), the phase-selection surface row (default, ticked, and never read as a phase), linked_work_items, render_assignment/render_graph_context surface + --pr-repo claim	uv run --directory cli pytest -q
T2	Integration (scenario)	yes	a cross-repo pull request routes to its work item, walks an inner loop whose state lands under the origin repo's spec directory, and holds the outer implementation gate until it completes	uv run --directory cli pytest -q cli/tests/test_graph_multirepo_integration.py
T3	Contract (OpenAPI)	yes	the authored contract gains an optional prRepo on the five graph bodies and the graphShow query, and the served schema still matches it route-for-route	uv run --directory cli pytest -q tests/test_api_contract_parity.py
T4	End-to-end	n/a — an end-to-end run needs a live GitHub app and two real repositories; T2 exercises the same seams against the shipped router, runtime and hooks with no network.
T5	UI / visual	n/a — no user-facing surface (CLI + markdown).
T6	Snapshot	n/a — no rendered output is snapshotted in this repository.
T7	Performance / load	n/a — the added work is one extra glob over a directory that holds one entry per pull request.
T8	Security / abuse case	yes	the four abuse cases of design.md § Security design: path traversal from a payload, from --pr-repo, an unarmed work item reached by a cross-repo link, and a declared repo that never gets a pull request	uv run --directory cli pytest -q -k "traversal or pr_repo or unarmed or declared"
T9	Accessibility	n/a — no UI.
T10	Migration / upgrade	yes	an existing work item's pr-loops/pr-<n>/ keeps resolving unchanged, and a graph-state.json with no surface field reads as the default (the work item)	uv run --directory cli pytest -q -k "back_compat or default"
T11	Manual exploratory	n/a — every behaviour is reachable from a test; nothing here needs a human to look at it.
T12	Parity (docs ↔ code ↔ schema)	yes	the new config key is in the schema, in READS, and in docs/config/harness-config.md; every CLI flag is documented	uv run --directory cli pytest -q cli/tests/test_docs_parity.py cli/tests/test_harness_config.py cli/tests/test_graph_parity.py
T13	Lint / type / format	yes	ruff, ruff format, pyright, markdownlint, and the harness-config schema validator	make lint typecheck (or the individual commands)

Scenarios & requirement trace

Row	Requirement(s)	Scenario / case
T1	R1.3, R1.4, R1.6	pr-loops/pr-7 for the origin repo, pr-loops/octo__infra/pr-7 for a contributing one; every malformed repo value raises
T1	R2.1–R2.5, R2.7, R2.9	the checklist offers the surface row; the surface resolves to work-item when untouched and pull-request when ticked; it is frozen into state and the frozen graph; an unticked row is neither a skip nor a refusal; the assignment and the prompt name it
T1	R1.5	a qualified closing keyword yields a ref in the other repository; an unqualified one stays local
T1	R4.1–R4.4	await-inner-loops: pass, wait-on-unfinished, wait-on-undeclared-loop, wait-on-declared-repo-with-no-loop, wait-on-unknown-origin
T2	R1.1–R1.5, R4.1	Scenario: a pull request in a contributing repository walks its own inner loop under the work item's spec directory
T2	R4.2	Scenario: the work item holds at implementation until every declared repository has finished
T8	abuse cases 1–4	negative tests named in design.md § Security design
T10	R1.4, R2.3	a pre-existing pr-loops/pr-<n>/ resolves unchanged; a state file with no surface reads as the default

Verification environment

• Repositories: this repository only. The multi-repo behaviour is exercised with fixture directories and synthesised payloads — no second checkout, no network.
• Services / containers: none.
• Fixtures & data: tmp_path trees written by the tests; webhook payloads built in-test from the shapes already used in cli/tests/test_poller.py and test_graph_loops.py.
• Credentials: none. No test in this work item authenticates to anything.
• Bring-up: uv sync --directory cli · Tear-down: none (pytest tmp_path).
• If bring-up fails: record it under Verification results, leave the dependent activities unticked, and escalate.

Evidence plan

Row	Evidence	Path under evidence/
T1, T2, T8, T10, T12	full suite output (counts, duration) and the red→green transitions	tests.md
T13	ruff / ruff format / pyright / markdownlint / schema-validator output	lint.md
T2	the scenario walk-through against the shipped graph and router	multirepo-scenario.md

Verification activities

• [x] T1 — uv run --directory cli pytest -q (unit)
• [x] T2 — uv run --directory cli pytest -q tests/test_graph_multirepo_integration.py
• [x] T8 — the four abuse-case tests, red before the mechanism existed
• [x] T10 — back-compat: existing pr-loops/pr-<n>/ and a config with no outerLoop
• [x] T3 — uv run --directory cli pytest -q tests/test_api_contract_parity.py
• [x] T12 — pytest -q tests/test_docs_parity.py tests/test_harness_config.py tests/test_graph_parity.py
• [x] T13 — ruff check, ruff format --check, pyright, markdownlint, validate_config.py

Verification results

Activity	Command / procedure	Outcome	Evidence
T1	uv run --directory cli pytest -q	pass — 1524 passed, 1 skipped	evidence/tests.md
T2	uv run --directory cli pytest -q tests/test_graph_multirepo_integration.py	pass — 6 passed	evidence/tests.md
T8	uv run --directory cli pytest -q -k "traversal or pr_repo or unarmed or declared"	pass — 68 passed; red recorded before the mechanisms existed	evidence/tests.md
T10	uv run --directory cli pytest -q -k "back_compat or default"	pass — 55 passed	evidence/tests.md
T3	uv run --directory cli pytest -q tests/test_api_contract_parity.py	pass — 1 passed (folded into the T12 run after the review round)	evidence/tests.md
T12	uv run --directory cli pytest -q tests/test_docs_parity.py tests/test_harness_config.py tests/test_graph_parity.py tests/test_api_contract_parity.py	pass — 32 passed	evidence/tests.md
T13	uv run ruff check cli hooks, ruff format --check cli hooks, pyright cli, markdownlint-cli2 "**/*.md", python scripts/validate_config.py	pass	evidence/lint.md
T2 (scenario)	scripted walk of the ticket's own scenario against the shipped router, runtime and hooks	pass	evidence/multirepo-scenario.md

Re-run after the PR #184 review round (the surface moved from the harness config to phase-selection): every row above is the re-run, plus pytest -q -k surface → 15 passed.

Not executed: none — every planned activity ran. One pre-existing tmux integration test (test_legacy_record_without_a_tmux_target_heals_via_respawn, untouched by this work item) failed once under the full suite and passed both in isolation and on a full re-run — a flake in this container, recorded here rather than smoothed over.

Review comments

Appended by the-loop's record-feedback hook when a human gate approves with comments.