[dmccoystephenson]

Summary

Stage B (unit-test expansion) cycle — characterization tests for the two parts of scripts/check_docs.py with the thinnest coverage. No production code is changed.

• extract_template_body (previously happy path + two missing-fence errors only): six new cases covering
  • backslash-escaped inner fences (\``bash`, the form the real template uses) not closing the template body
  • a fence with an info string (```bash) not closing the body
  • an indented ``` not closing the body (closing fence must be exactly ``` at column 0)
  • the opening line being strip()ped before comparison, unlike the closing line
  • a closing fence with trailing whitespace not being recognized (reported as a missing closing fence)
  • the first ```markdown fence winning when there are several
• main() (previously untested): new MainTests class, with REPO_ROOT / TEMPLATE_FILE / README_FILE pointed at a temp directory and restored in tearDown
  • consistent docs → exit 0 and the "passed" line
  • one failure from each of the three checks → exit 1, all three printed under "FAILED"
  • a missing template fence → the placeholder check is skipped, so only the fence error is reported

All assertions describe current behavior. None of the cases turned up a bug that needs a fix.

No tracking issue — gap found during triage.

Deferred backlog: every other open issue was left for a later cycle. Most of them (#123–#158, labeled template-rule) change the generated-skill template. Their only behavioral anchor is a real /create-dev-loop run, which would create a GitHub repo and write outside the checkout, so a headless dispatch cannot verify or auto-merge them. The other open issues (#106, #109, #114, #128, #133, #135, #141–#143) would need either CLAUDE.md edits, which the harness blocks, or decisions about process/design that a maintainer should make.

Doc sync check

Not applicable — only tests/test_check_docs.py changed.

Test plan

• python3 -m unittest discover -s tests -v — Ran 41 tests ... OK locally (32 before this PR, 9 added)
• python3 scripts/check_docs.py — Doc consistency check passed.
• CI doc-consistency green on the PR head

This PR description was drafted during a Gardener session (https://github.com/Stephenson-Software/gardener).

🤖 Generated with Claude Code

---

drafted by Claude on behalf of Daniel Stephenson

[dmccoystephenson]

Self-review rubric (posted as a plain comment; the review was performed inline during this headless run):

• Scope: PASS — the diff touches only tests/test_check_docs.py (+91 lines). No production code, docs, or formatting churn.
• Tests-new: PASS — no new public functions were added. The diff adds cases for extract_template_body (6) and for main() (3, a function that had no tests until now).
• Tests-fix: N/A — this is a characterization-only PR with no bug fix. To show the new cases can fail, the closing-fence comparison in extract_template_body was temporarily changed to lines[i].strip().startswith("```"). Three of the new tests then failed (test_closing_fence_with_trailing_whitespace_is_not_recognized, test_fence_with_info_string_does_not_close_template, test_indented_fence_does_not_close_template). After git checkout -- scripts/check_docs.py, all 41 tests passed again.
• Sibling structure: PASS — MainTests follows the existing pattern. It has one TestCase class per function under test, resets check_docs.errors = [] in setUp(), restores the patched module globals in tearDown() the same way CheckLocalLinksTests restores REPO_ROOT, builds its fixtures inline in a tempfile.TemporaryDirectory, and sits last so the class order matches the source file.
• Sibling renames: N/A — nothing was renamed.
• Docs: PASS — the module docstring's statement that "any new test class added here must do the same in its own setUp()" still holds for MainTests. No other doc describes the test inventory.
• Issue resolution: N/A — no tracking issue. This is a Stage B cycle, as the PR body states.
• CI: PASS — doc-consistency is green on the PR head. The CI log shows Doc consistency check passed. and Ran 41 tests (it was 32 on main).
• Placeholder parity / README Step parity / Fence escaping / Phase numbering / Version comments: N/A — create-dev-loop.md and README.md are not touched.
• Research grounding: N/A — the template is not touched. Stage B rests on RESEARCH.md §3 (regression guards).
• No back-ported specifics / gardener interface / Retrofit note / Behavioral validation: N/A — no template or Step changes.

Judgment call noted, not blocking: test_closing_fence_with_trailing_whitespace_is_not_recognized pins down an asymmetry. The opening fence is strip()ped but the closing fence is not, so a closing fence with trailing spaces currently reads as "missing closing fence". The test records that behavior and does not endorse it. If the asymmetry is ever treated as a bug, this test is the one that should flip.

This comment was drafted during a Gardener session (https://github.com/Stephenson-Software/gardener).

---

drafted by Claude on behalf of Daniel Stephenson