[dmccoystephenson]

Summary

• A Prove each guard can fail bullet is added to the generated template's Phase 2 Stage B (unit-test expansion). For at least one new test per target, a small, plausible mutation of the characterized production code is made temporarily, the test is confirmed to fail, and the file is restored with git checkout --. A test that survives every plausible mutation is to be strengthened or dropped, and the mutation is never committed.
• A matching Tests-mutation (empirical, not judged) item is added to the Phase 4 universal rubric, directly after Tests-fix. Tests-fix only covers bug fixes, so a Stage B PR previously scored it N/A and nothing checked that the new tests could fail at all. The new item is N/A for PRs that add no characterization tests.
• This writes down the step that was done by hand in PR Add extract_template_body edge cases and main() tests #159, where three of nine new tests were confirmed to fail against a mutated closing-fence comparison in scripts/check_docs.py.
• No placeholder is added and no Step is added or removed, so neither the Step 4 table nor README's "What it does" list changes. No phase is renumbered. Steps 3, 5, and 6 are untouched, so nothing changes for gardener.

Closes #160

Research grounding

RESEARCH.md §3: regression evidence is what tells a real guard apart from a coincidental one. Tests-fix already applies that to bug fixes, and this change applies it to characterization tests written without a fix. An Implementations entry is added under §3. Its ship date is written as the authoring date, which has the known problem described in #128.

Retrofit

Generated loops that predate this change have a Stage B with no mutation check. Any loop that has run a Stage B cycle is likely to need a retrofit pass, since those loops are the ones that score Tests-fix as N/A. This includes create-dev-loop-dev-loop, which ran one for #159. Per CLAUDE.md's "Promoting a rule into the template", the retrofit is left to a follow-up pass.

Doc sync check

• RESEARCH.md updated (an Implementations entry under §3)

Test plan

• python3 scripts/check_docs.py: "Doc consistency check passed."
• python3 -m unittest discover -s tests: ran 41 tests, OK
• Neither added line contains a fence. The count of unescaped ^```` lines in create-dev-loop.md` is unchanged.
• UNVERIFIED (behavioral): no /create-dev-loop run was performed. This headless dispatch is not authorized to create a GitHub repo (Step 6) or to write to ~/local-skills/ / ~/.claude/commands/ (Steps 3/5). None of the five CLAUDE.md "Testing changes" items were exercised. For that reason the PR is left for a maintainer to merge, not auto-merged.

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 comment in this headless dispatch, standing in for a formal review):

• Scope: PASS — the diff is 2 template lines in create-dev-loop.md (one Stage B bullet and one Phase 4 rubric item) plus 1 RESEARCH.md Implementations line. All three are needed for Stage B has no check that new characterization tests can fail #160.
• Tests-new: N/A — no function was added. The only code is scripts/check_docs.py, which is unchanged.
• Tests-fix: N/A — this is not a bug fix.
• Sibling structure: PASS — no new file. The new rubric item follows the **Name (empirical, not judged):** form of Tests-fix, and the Stage B bullet follows the bold-lead style of the bullets around it.
• Sibling renames: N/A — nothing was renamed.
• Docs: PASS — the Phase 7 table was checked. README, CONTRIBUTING.md, SECURITY.md, the PR and issue templates, and scripts/README.md do not restate the rubric items or Stage B's bullets (searching for Tests-fix and Stage B outside create-dev-loop.md finds only RESEARCH.md). RESEARCH.md §3 has the new Implementations entry. CLAUDE.md is unaffected.
• Issue resolution: PASS — Stage B has no check that new characterization tests can fail #160 asked for a Stage B step and a rubric line next to Tests-fix. Both are added, with the issue's suggested wording kept apart from small edits for the template's voice.
• CI: PASS — doc-consistency is green on the PR head (7s).
• Placeholder parity: PASS — no placeholder was added. check_docs.py: "Doc consistency check passed."
• README Step parity: N/A — no Step was changed.
• Fence escaping: PASS — no fence was added. The count of unescaped `^```` lines is the same as before (28).
• Phase numbering: PASS — no phase was renumbered. The new text refers to "Phase 2 Stage B" and "Phase 4 rubric", and both still exist.
• Version comments: N/A — the header is not touched.
• Research grounding: PASS — §3 is cited, and a PR Require Stage B characterization tests to be shown failing against a mutation #162 Implementations entry has been added.
• No back-ported specifics: PASS (borderline) — the PR body refers to scripts/check_docs.py only as the PR Add extract_template_body edge cases and main() tests #159 example. The template text itself names no repo-specific file or command.
• gardener interface: N/A — Steps 3, 5, and 6 are unchanged.
• Retrofit note: PASS — the PR body says which loops likely need a retrofit (those that ran Stage B, including create-dev-loop-dev-loop).
• Behavioral validation: FAIL (UNVERIFIED) — no /create-dev-loop run was performed. The headless dispatch is not authorized for Step 6 repo creation or for writes outside the checkout. For that reason the PR is not auto-merged.

Judgment call for the maintainer: the Phase 4 step 5 example comment body still lists only Tests-new/Tests-fix before its .... This was left as is because the ... already stands for the remaining items.

Summary: the change is mechanically consistent and CI is green, but the template's behavior has not been validated. It is ready for a maintainer to review and merge.

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

---

drafted by Claude on behalf of Daniel Stephenson

[dmccoystephenson]

Triage skip reasons for this cycle (one PR per cycle; #160 was selected because it is small, well specified, and backed by an observed instance in PR #159):

• CLAUDE.md still links gardener under the old dmccoystephenson owner #106: needs an edit to CLAUDE.md, which is agent-loaded config that the loop may not edit without separate authorization. Left for the maintainer.
• RESEARCH.md Implementations entries record a ship date that is unknowable when the entry is written #128: the issue offers convention options and asks for a maintainer decision. The fix also needs a CLAUDE.md edit.
• [self-audit] The repo's own validation anchor is structurally unrunnable in a headless dispatch, so template changes can never merge autonomously #109, Unverified: whether an in-checkout .worktrees/ survives gardener's refresh (git clean -fdx) #114: tracking or verification issues about the headless environment, not doc or template changes this loop can make.
• Phase 4's rubric scores the diff only, so PR-description obligations like CLAUDE.md's retrofit note go unchecked #120, gh prose examples carry no attribution sign-off, and --body-file paths must be literal (a shell variable is invisible to the hook) #122, do-not-auto-merge path rules conflict with a run-level --allow-merge grant, with no stated precedence #123, The >50-deleted-lines do-not-auto-merge heuristic misfires on extraction refactors, test migrations, and generated artifacts #124, Negated closing keywords in PR bodies still auto-close issues #125, grep is not on tend's allow-list, so generated validation checklists and verify-by-removal assertions cannot run headlessly #127, An approval-gated CI run reads as 'no checks reported', which is neither the green nor the red Phases 1 and 4 branch on #129, A restricted fetch refspec also blocks checking out an existing remote branch, with an error that reads as 'branch does not exist' #130, Test the mechanism, not the stochastic outcome, when a mock cannot exercise an acceptance criterion #131, Phase 4 verifies backward 'Closes #N' references but not forward 'a follow-up issue has been filed' promises #133, Phase 5's read-only 'gh api .../pulls/<n>/comments --jq' is refused by the attribution hook as an unsigned post #135, Tests-fix has no path for a scope-blind anchor, only for an anchor that cannot run #140, Upstream COMMANDS.md is trusted as source, but it can disagree with the command class it documents #141, A blanket merge pre-authorization does not resolve a do-not-auto-merge path match, and Phase 8 does not say so #142, Stage A sweeps fix the instances named, not the class — and a prose grep is treated as a CI-strength anchor #143, Phase 1 triage never checks whether the default branch's own CI is green #144, A refused environment probe is not grounds for UNVERIFIED — the anchor has to be attempted first #145, Phase 2: bundle a small docstring/annotation fix with a co-located test or doc update as one PR #146, Phase 3: never pre-escape Edit-tool strings — verify the diff for literal backslash-quotes after inserting quoted text #147, Phase 1 repo resolution assumes the configured clone is clean; stale staged work in a shared cache clone blocks git pull #148, Tests-fix revert produces compile errors, not failures, when the new tests reference symbols the fix introduces #149, Phase 5: compress the review wait for a single autonomous run with no reviewer assigned or requested #150, Dispatched runs can load an installed skill copy older than the rules merged in the skill's repo #158: template-rule or process changes. They were deferred to keep the cycle to one coherent PR. Like this one, each is a template-behavior change that cannot be validated by a headless run, so batching them would only widen the review the maintainer has to do.

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

---

drafted by Claude on behalf of Daniel Stephenson