chore,docs: track .claude/agents, delete the axis twin, add review axis 11

[derek73]

Closes #473.

.claude/agents/design-docs-reviewer.md implements docs/design/AGENTS.md's review axes and was untracked, so nothing diffed it and nothing reviewed it. #473 reports one missing axis; measured, the drift is three-way and one-directional — every correction landed in the tracked file and none reached the untracked copy:

	agent file	docs/design/AGENTS.md
axis 4	inward only ("Rule vs. the decision entries it cites")	inward and outward
axis 10	absent	"Re-run every capability claim."
corpus caution	"exactly one ends in a particle"	"two of those end in a particle" — recorded there as the corrected count

Four commits

1. 35542d0 — un-ignore .claude/agents/ and .claude/commands/, adding both files byte-unchanged. .claude/ with a trailing slash makes git stop descending, so the negation needs .claude/* first; verified that worktrees/, settings.local.json, launch.json and .DS_Store stay ignored.
2. 9f1d46e — the agent drops all four duplicated sections (−104 lines) and points at AGENTS.md. Because commit 1 added the file unchanged, this diff is the evidence: it removes ## The nine axes, the narrowed axis 4, and the pre-correction corpus count.
3. dc17090 — axis 11, plus the axis-count claims it invalidates.
4. 6b6e297 — fixes ten findings from two review agents, all of them in commit 3. See below.

No pin test

#473 asks for a test pinning the agent's axis headings against AGENTS.md's list. Not added, and decisions.md#review-agent-single-source records why: heading comparison would have caught the missing axis 10 and neither of the other two drifts, since both changed body prose and left headings intact. The twin is the defect; a pin test detects a third of it and licenses keeping the rest. Same shape as the STABLE_TAGS twin flagged in nameparser/_types.py.

Axis 11 — state the invariant that relates two DIFFERENT inputs

Every existing axis compares a document to another document, to its own examples, or to the implementation, so all ten can only find code that disagrees with a rule — none can find a rule that is simply wrong. That is what backed out #461's fix and closed #466 unmerged.

The worked invariant: the family-comma writing under the default order and the comma-less writing under FAMILY_FIRST must parse alike.

"Vega, Juan de"  → family "de Vega"    "Vega Juan de"  → family "de Vega"                ✓
"Vega, Juan vd"  → family "vd Vega"    "Vega Juan vd"  → family "Vega", suffix "vd"      ✗

Measured over the seven role fields, PARTICLES × 6 families × 6 givens: it holds on all 36 pairs for 67 of the 70 particles and fails on all 36 for do, mc and vd — precisely the words that are also suffix vocabulary, which rules.md#P6's Accepted consequence already names as the contested memberships. The invariant found that set from behavior, having read no rule.

The measurement is bounded: families and givens must not themselves be particle vocabulary. With F="Van" it is 0 of 70, not 67 of 70 — the spread word joins the particle run and the invariant measures a different shape. decisions.md#two-input-invariants carries this and the recompute recipe. The axis prose itself carries no bare count, per the counting-claims rule.

What review changed (commit 6b6e297)

Two agents reviewed commit 3 — the project's own design-docs-reviewer and a general convention reviewer. Ten findings accepted, one rejected. The three that mattered shared one cause: the new entry re-derived arguments decisions.md#P6 already makes, instead of citing them — the duplication this PR exists to delete.

• The Declined: block claimed "216 of 216" came from "six hand-picked particles". decisions.md#P6 already records the real provenance — a sorted(particles)[:14] slice — already calls it axis 2 in its purest form, and already closes "Recompute over the WHOLE vocabulary, never a sample." The block now cites that entry instead of contradicting it.
• It attributed fix(post_rules): a trailing particle attaches to the family under a declared family-first order (#365) #466's failure to do/mc/vd. design-docs-reviewer implements nine of AGENTS.md’s ten axes, and the missing one would have caught #467’s false premise #473's "ambiguous-particle half" is the 37 non-never-given particles; decisions.md#P6 records that fix(post_rules): a trailing particle attaches to the family under a declared family-first order (#365) #466 died on a never-given test excluding all 37, van among them. Different partition, different failure.
• The recompute recipe said "any spread of families and givens" — false, as above.
• AGENTS.md's corpus caution loses its standing count: both the drifted and the "corrected" wording quoted figures decisions.md#P6 had already superseded (measured today: 305 of 1104 names carry a comma, 13 end in a particle).
• Axis 11 is no longer classed rules.md-specific — that contradicted its own "consults no rule statement" and would have had a reviewer skip it on a decisions.md-only change like this one.
• Axes 10/11 no longer said to come from "later arcs" (Recognize multi-word credential suffixes after a comma ("Smith, LEED AP") #291 predates docs(design): the three keystone decisions from the post-#381 backlog review #386); root pointer drops its numeral; docs-review.md stops naming a section commit 2 deleted; the agent's description now names AGENTS.md, which the axes bind and it omitted.

Rejected: rephrasing "#461 was backed out" to "the fix for #461" — accurate, but decisions.md:746 and :754 already use that shorthand, and fixing one of three sites makes it inconsistent.

Test plan

• uv run --frozen pytest -q — 6198 passed, 156 skipped, 9 xfailed
• git check-ignore over eight .claude/ paths — 2 trackable, 6 ignored
• axis count agrees across docs/design/AGENTS.md heading, its numbered list, and the 3+8 split
• invariant re-measured on the committed tree by me and independently by both review agents — identical
• git diff master...HEAD --stat — six files, no docs/superpowers/ path

The executable form of the invariant stays with #469.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.66%. Comparing base (c34fb74) to head (6b6e297).

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #481   +/-   ##
=======================================
  Coverage   98.66%   98.66%           
=======================================
  Files          45       45           
  Lines        3218     3218           
=======================================
  Hits         3175     3175           
  Misses         43       43           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.