Testing plan: everything can be declared away, nothing is declared away quietly

Phase 3 of 4. Derived from the locked requirements.md and design.md. Ticket: issue #179.

This file is executable content. It names commands an agent will run, so review it like code. No credentials are involved: every test runs against temp directories and this repository's own checkout; nothing touches the network.

Test matrix

#	Type	In scope?	What it proves
M1	Unit — the vocabulary is exactly right	yes	R1.1, R1.2: every outer-loop node is skippable except phase-selection (still required: true) and the terminals; asserted as a set equality so a future node cannot be added silently on either side
M2	Unit — the invariant	yes	R1.2, R1.3: phase-selection is required and not skippable; security-review/human-approval are skippable and no longer required; required × skippable is still a compile error (the rule that enforces the invariant)
M3	Unit — routing is authored	yes	R1.4: the shipped graph compiles (every skippable node declares its on: skipped edge), and each new edge points at the node's ordinary forward successor
M4	Unit — skip sets	yes	R1.5: spec-chain names the seven spec-chain nodes; review-chain names the six review nodes; every member is skippable (compile-checked)
M5	Unit — the inner loop is untouched	yes	R1.6: pdlc-pr-loop declares no skippable node, has no phase-selection, and keeps security-review required: true
M6	Unit — routing end to end	yes	R1.8: with every selectable phase declared away the pointer runs phase-selection → complete; each intermediate node records outcome skipped; none of their hooks run (no phase label, no log entry, no assignment)
M7	Unit — onlyWhenSkipped applies	yes	R3.1: the entry gates its fallback artifact when the named artifact is a planned absence (authoring node declared-skipped and absent)
M8	Unit — onlyWhenSkipped does not apply	yes	R3.1–R3.3: skipped-but-present → does not apply; not skipped → does not apply; names an artifact no skippable node authors → never applies; parameter absent → today's behaviour unchanged
M9	Integration — the real verification node, plan kept	yes	R2.1: testing-plan.md locked and complete → pass; its Verification results empty → block. No execution-log section required
M10	Integration — the real verification node, plan declared away	yes	R2.2: no testing-plan.md and test-planning skipped → blocks naming execution-log.md; once the log carries non-empty Verification results → passes
M11	Integration — declared but present	yes	R2.3: test-planning skipped and a testing-plan.md present → gated on the plan alone; an empty log section does not block
M12	Unit — the checklist copy	yes	R1.7: with no protected phase the checklist says every phase is selectable and each omission is recorded, instead of printing an empty "always runs" block; the selectable rows still name every skippable node
M13	Parity — templates, phases, manifest	yes	R2.5: test_p5c (every gated section exists in the artifact's template) passes, which it can only once templates/execution-log.md offers Verification results; P1–P5b unchanged
M14	Regression — full suite, lint, types, docs	yes	nothing else moved: pytest, ruff, ruff format --check, pyright, markdownlint, validate_config.py
M15	UI/visual, accessibility	n/a	no user-facing surface — YAML, Python and markdown only
M16	Performance	n/a	one extra set-membership test per gated node; ten more edges in a ≤ 20-node graph. Nothing hot
M17	Migration	n/a	additive: an already-frozen graph keeps walking the phases it froze (a skip is recorded, never inferred), and an in-flight item sees the wider checklist only at a selection it has not yet answered
M18	E2E against live GitHub	n/a	no integration surface changed — the selection gate, its transports and the CLI verb are untouched

Verification environment

This repository's own checkout, nothing else: uv for the environment, pytest from cli/, the linters from the repo root — the same commands CI runs (uv run --directory cli pytest -q, uvx ruff check, uv run --directory cli pyright, npx markdownlint-cli2). No second repo, no service, no secrets; integrations are faked in-process.

Evidence plan

Committed under evidence/:

• tests.md — the new tests red against the pre-change code (proving they test something), then green, with full-suite counts before and after.
• walkthrough.md — the motivating scenario end to end: a doc-fix work item declaring spec-chain + review-chain, the-loop check reporting each omission with provenance, and verification blocking until the execution log carries its results.
• lint-and-types.md — ruff, pyright, markdownlint and config-validation output.

Activities checklist

• [x] M1–M13 written test-first and failing against the pre-change code where the behaviour is new (record the red in evidence/tests.md)
• [x] M1–M13 green
• [x] M14 full regression suite + lint + types green
• [x] Evidence committed and redacted (no tokens, no hostnames beyond github.com)

Verification results

Executed at the verification node on 2026-08-08. Per-activity record below; raw output in evidence/.

Activity	Command	Outcome	Evidence
M1–M6, M12 vocabulary, invariant, routing, checklist	uv run --directory cli pytest -q tests/test_graph_skips.py	pass — 46 tests. Red first: five of them failed against the pre-change YAML and checklist (ten nodes unmarked, security-review still required, no review-chain set)	evidence/tests.md
M7, M8 onlyWhenSkipped	uv run --directory cli pytest -q tests/test_graph_hooks.py	pass — 46 tests, four of them new. The three dormancy cases were red before the parameter existed (the entry applied unconditionally)	evidence/tests.md
M9–M11 the real verification node	uv run --directory cli pytest -q tests/test_graph_verification_integration.py	pass — 10 tests, three of them new; the plan-declared-away case blocked naming execution-log.md before the section was written, then passed	evidence/tests.md
M13 parity	uv run --directory cli pytest -q tests/test_graph_parity.py	pass — 8 tests; test_p5c red in isolation before templates/execution-log.md offered Verification results, exactly as designed	evidence/tests.md
M1/M4 the operator surface	uv run --directory cli pytest -q tests/test_core_graphs.py	pass — 5 tests; the shipped-vocabulary case now declares seven spec-chain nodes and rejects phase-selection rather than security-review	evidence/tests.md
M14 full suite	uv run --directory cli pytest -q	pass — 1480 passed, 1 skipped (baseline 1467 passed, 1 skipped)	evidence/tests.md
M14 lint/types/config	uvx ruff check cli · uvx ruff format --check cli · uv run --directory cli pyright · npx markdownlint-cli2 over the changed docs · uv run python scripts/validate_config.py	pass — all clean	evidence/lint-and-types.md
Walkthrough (the ticket's scenario)	the-loop graph skip + a scripted selection-gate run against the shipped graph, then check and the verification node	pass — 13 nodes declared from two tokens with phase-selection refused, pointer at implementation, every omission reported skipped by declaration — by whom, verification blocked until the log carried its results, and a forged declaration on the gate inert and surfaced	evidence/walkthrough.md