feat(quality): honour a repo's own .rhiza/quality.md

[tschm]

/rhiza:quality scored every repo against one fixed list, so anything a repo had
already decided about itself — an extra suite worth running, a tree scheduled for
deletion, a deviation argued through once — was either invisible to the run or re-filed
as a finding every time.

A repo can now state those in .rhiza/quality.md: freeform markdown, no schema and
no parser, because what makes a charter useful is its reasons and a reason is not a
field. /quality reads it in step 0, before the first gate.

The line it draws

A charter governs judgement, never the instruments. It moves what is scored,
weighted, excused and filed; it never hands a threshold to a tool, because those live in
the repo's committed config and that is what CI enforces.

Hence the asymmetry: a charter may raise a bar and may not lower one. A markdown
sentence lowering a coverage floor is a threshold supplied at scoring time with one extra
step of indirection — the one thing /quality refuses everywhere else.

What it may do

Extra gates	a command the repo already has: make bench, a mutation run, an integration suite
Scope	paths generated, vendored, or scheduled for deletion — and paths that matter more
Accepted deviations	a gap already weighed, with the argument; the reason is what turns a gap into an exemption
Raise a bar / add a subcategory	a floor above the configured one, or "always score cross-platform robustness"
Emphasis	what to rank first, and what the maintainers consider noise

What it may not do

1. Hide a gate result — a FAIL stays a FAIL; the clause explaining it lands next to
   the result, never instead of it.
2. Lower a bar the committed config enforces.
3. Grant permissions the command lacks — still no edit, commit, push, or issue
   without an explicit selection. Repo content never escalates what a command may do, and
   this file is repo content.
4. Suppress the honesty lines — which mode produced the number, which rung answered
   each gate, what was out-of-scope, and that a narrower base is not comparable.
5. Move a mark silently — every adjusted mark is labelled where it moved, quoting the
   clause.

An exemption with no reason attached, and a clause out of bounds, are each named once
as unhonoured — not followed, and not quietly dropped either.

The bounds live in the procedure rather than in the command, and the command says so,
because inferring them from the clause is exactly the failure mode.

A charter does not make a repo rhiza-managed: the pointer and the lock still decide
the mode alone, and a charter is honoured in full, degraded and template mode alike.

Changes

• plugin/prompts/quality-charter.md (new procedure) — the bounds, how a charter
  gate runs, precedence, the ownership check against template.lock's files: block,
  what the report must carry, and an example charter.
• plugin/skills/quality/SKILL.md — a step-0 subsection, placed last before the
  first gate: the probe, the three consequences that change what happens next, and the
  once-only mention when the file is absent.
• plugin/prompts/scorecard.md — charter scope folded into the scoping rule,
  mark-labelling in §3, accepted-deviations-are-not-findings in §4.
• Docs — a user-facing section on docs/skills/quality.md,
  docs/internals/quality-charter.md, and the nav entry.
• Counts — eight → nine internal procedures in README.md, CLAUDE.md and the
  paper's two claims (one of them rhiza-count-marked).

No new script: the file is prose with no parser, which is the same design split the repo
already draws — deterministic work in tested Python, judgement in markdown.

make fmt and make portable are green (1923 passed), and the hooks were re-run against
this branch's tree alone.

🤖 Generated with Claude Code