Let the measurement liturgy have one copy, and give the section its budget back

[MarkusNeusinger]

Follow-up to #540, which had to write the same --expect-root flag into two places and said so in the budget comment it added.

The duplication

The five-step measurement liturgy of a Tintenfolger round stood twice:

• .claude/skills/verify-trace/SKILL.md — the executable form, loaded when the skill is invoked;
• docs/reference/werkzeuge.md § „Benches und Generator" — the same five steps as copy-paste command blocks, ~50 lines.

#540 changed both. Its own budget comment named the fix and deferred it:

The real fix is older than this raise and belongs to its own PR: those five steps are a second copy of .claude/skills/verify-trace/SKILL.md (both had to be edited with the same flag here), and collapsing them to a pointer would give the section back ~700 tokens and remove the drift risk with them.

What the doc keeps, and what it points at

werkzeuge.md is the inventory — what exists, what it is called, which invariant hangs on it — so that is what stays:

• the five entry points of tools/tracebench and tools/pairlab/follow (follow, tracebench, .k0eval, .excursions, .view), each with one clause on what it produces;
• where the flags are documented: each tool's --help, and each arm in its own §14 entry — that is where they were already authoritative, the doc only copied them;
• --expect-root on all five, and the invariants (pure measurement layer, dev split frozen and append-never, paired comparisons only inside one pinned environment and one root).

In its place, one paragraph on the skill: what it covers, that it is invoked before a round rather than read afterwards, why that matters (aug25 L-U and aug26 v5 were both measured against the wrong follower, from memory), and that the doc stays the inventory while the round's sequence has exactly one source. The Stand block carries the same in one line, since a session reads it instead of the file, and its „Was offen ist" no longer claims the tools have no skill — the trace round now has one.

The budget

	vorher	nachher
§ „Benches und Generator" (the widest section, and therefore the measured one)	4125	3704
werkzeug-abschnitt budget	4538 (#540)	4073
headroom	+413	+369

3704 is two tokens off the 3702 that set the 4073 budget on 2026-09-04 — the section is back where it was before the flag lines went in. So the raise is reverted to the same number, not re-derived from the new measurement: the growth was moved out of the read path, which is what the gate asks for, and a budget that re-measures to its original value is the proof that it was a stopgap. uv run python -m tools.docs_budget check green, every path inside budget.

Pointers checked

• docs/index.md — the werkzeuge.md row describes the tool index and needs no change; nothing in the map referenced the liturgy.
• docs/reference/glossar.md — the k0eval entry pointed at „werkzeuge.md (Mess-Liturgie)"; it now names /verify-trace for the liturgy and werkzeuge.md for the inventory.
• .claude/skills/verify-trace/SKILL.md — its opening said the liturgy is one „that werkzeuge.md records". It now says this file is its only copy, and why.
• The dated §14 entries in messjournal.md that mention „Mess-Liturgie (werkzeuge.md)" are left as written — closed records are not rewritten, and the pointer under that same heading still leads to the skill.
• CLAUDE.md / .github/copilot-instructions.md — neither mentioned the liturgy; unchanged, and tests/test_agent_instructions.py is green.

Gates

tools.docs_budget check, tools.docs_register check, tools.changelog check, ruff check + format --check, full backend suite (2433 passed / 18 skipped) — all green locally.

🤖 Generated with Claude Code

https://claude.ai/code/session_01UEScQMZFvxxNNyNJYryfa3

[Copilot]

🔵 Needs a closer look

Two documentation nits remain unresolved, and approval-readiness assessments are split.

Pull request overview

Consolidates the Tintenfolger measurement procedure into /verify-trace and restores the documentation budget.

Changes:

• Removes duplicated measurement commands from the tool inventory.
• Makes /verify-trace the canonical procedure.
• Updates budget, glossary, and changelog references.

File summaries

File	Description
tools/docs_budget/__init__.py	Restores the budget; wording incorrectly calls 4073 the measured size.
docs/reference/werkzeuge.md	Retains the inventory and points to the skill; includes an inaccurate skill-coverage claim.
docs/reference/glossar.md	Separates procedure and inventory references.
changelog.d/liturgy-pointer.md	Records the documentation consolidation.
.claude/skills/verify-trace/SKILL.md	Establishes the canonical measurement procedure.

Review details

Suppressed comments (2)

docs/reference/werkzeuge.md:33

• The claim that only the Tintenfolger round has a skill is inaccurate: .claude/skills/optimize-glyphs/SKILL.md:3,8-10 defines an experiment skill around tools/glyphbench. Keep the intended contrast without making the inventory's skill coverage incomplete.

> die Prüfung ist der Bench-Lauf und das Auge; einen Skill hat nur die
> Tintenfolger-Runde (`/verify-trace`). Die Methode und die Zahlen eines

tools/docs_budget/init.py:237

• 4073 is the restored budget, not the measured section size—the comment below correctly says the section measures 3704 (and the original measurement was 3702). Calling 4073 “measured” makes the rationale for this magic number internally contradictory.

# `werkzeug-abschnitt` was raised to 4538 on 2026-09-05 and is back at its
# measured 4073 the same day. The raise was honest but a stopgap: „Benches und

• Files reviewed: 5/5 changed files
• Comments generated: 0
• Review effort level: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.