Testing plan: write the-loop's artifacts for a human reader

Derived from requirements.md and design.md. Authored at test-planning, completed at verification.

This file is executable content. Review the commands like code. No credentials are involved — every command reads this repository and nothing else.

Test matrix

#	Type	Applies?	Scope / what it proves	Where it runs
T1	Unit	yes	test_writing_parity.py P1–P4: skill present and parsing, every human-read template points at it, the pointer names the skill the schema declares and no length limits have returned, no P0 tell in shipped prose	make test
T2	Integration (scenario)	n/a — the change adds no runtime path. Nothing is dispatched, routed or spawned; there is no cross-module behaviour to document with a Gherkin scenario.
T3	Contract (OpenAPI / GraphQL SDL)	n/a — no API surface added. docs/api-specs/openapi is untouched.
T4	End-to-end	n/a — no user-invocable flow. The skill is read by an agent, not executed.
T5	UI / visual	n/a — no user-facing surface (design.uiArtifacts produces nothing for docs/CLI work).
T6	Snapshot	n/a — no rendered output to freeze.
T7	Performance / load	n/a — filesystem reads in a test; no runtime cost.
T8	Security / abuse case	yes	the abuse cases in design.md §Security design: a style pass cannot rewrite a record (P4's glob excludes docs/specs/ and evidence/), and a missing or wrong-skill pointer fails rather than skips	make test
T9	Accessibility	n/a — no UI.
T10	Migration / upgrade	yes	a project scaffolded before this change has no writingStyle block; schema defaults must make absence and default the same state	make validate
T11	Manual exploratory	yes	read the rendered SKILL.md and one human-read template as a reviewer would, and confirm the contract is findable without reading the spec	manual

Scenarios & requirement trace

Row	Requirement(s)	Scenario / case
T1	R1.1, R1.3	the writing skill exists, front-matter parses, reference/tells.md present
T1	R2.1	every human-read template names the governing skill
T1	R2.1, R5.1	the pointer names the skill writingStyle.skill declares
T1	R2.2, R5.1	writingStyle.budgets is absent — length limits cannot return unremarked
T1	R5.2	no P0 tell in shipped prose — skills/, commands/, rules/, README.md and docs/ minus the historical (docs/specs/) and generated trees
T8	R2.4, abuse case 2	human-read templates still carry their gated sections
T8	abuse case 1	P4's scan excludes docs/specs/ and evidence/
T10	R5.1	.the-loop/harness-config.yaml and the shipped template both validate against the schema

Verification environment

• Repositories: this repository only.
• Services / containers: none.
• Fixtures & data: none — the test reads checked-in files.
• Credentials: none. No command here touches a secret.
• Bring-up: make install-dev · Tear-down: none.
• If bring-up fails: record it under Verification results, leave the dependent activities unticked, and escalate.

Evidence plan

Row	Evidence	Path under evidence/
T1, T8	pytest summary for the new test plus the full suite	unit.md
T10	make validate output (both configs against the schema)	validate.md
all	make lint + make typecheck + make format-check output, plus the-loop's own gate	checks.md
T11	the manual read-through finding	Verification results, below

Verification activities

• [x] T1 — uv run --project cli python -m pytest -q cli/tests/test_writing_parity.py
• [x] T1/T8 — make test (full suite: no regression from the schema and template edits)
• [x] T10 — make validate, plus the absent-block and rejected-key cases
• [x] all — make lint, make format-check, make typecheck
• [x] T11 — read skills/writing/SKILL.md and skills/the-loop/templates/design.md as a reviewer; record whether the contract is findable without the spec

Verification results

Activity	Command / procedure	Outcome	Evidence
T1	uv run --project cli python -m pytest -q cli/tests/test_writing_parity.py	21 passed	evidence/unit.md
T1/T8	make test	1349 passed, 1 skipped	evidence/unit.md
T10	make validate + a config with writingStyle removed, an unknown key under writingStyle, and a typo'd formal register	both configs valid; the pre-issue-165 shape still validates; the unknown key and the typo'd register are both rejected	evidence/validate.md
all	make lint, make format-check, make typecheck	ruff clean · markdownlint 0 errors over 420 files · pyright 0 errors	evidence/checks.md
T11	read SKILL.md and templates/design.md as a reviewer	the contract is visible where the artifact is authored (first lines of the template) and names the skill governing it; no need to open the spec	this table
—	the budget experiment that preceded the current design	recorded for the record: it is what showed the numbers to be unworkable	evidence/budgets.md

T11 finding, and what it cost the design. Verification measured every artifact this PR ships against the budgets it then proposed, and three of them did not hold: tasks: 200 was unreachable from its own 274-word empty template, requirements.md ran 682/500 and design.md 1017/900. The numbers were corrected and a sixth assertion added to keep budgets reachable — and then the owner rejected budgets outright on PR #168, for the underlying reason those corrections were evidence of. Length limits are gone; the record of the experiment is evidence/budgets.md and decision-061 §D2.

Not executed: none. Every in-scope activity ran, and re-ran after the budgets were removed.

Review comments

Appended by the-loop's record-feedback hook when a human gate approves with comments (issue-109).