refactor(code-metrics,mutation-testing): read team settings only from docs/conventions/<concern>.yaml #12290

Parent

Refs #5906

Outcome

Neither plugin reads a team setting from a .claude/ file:

• code-metrics reads its team layer only from docs/conventions/code-metrics.yaml. Its read of the older
  .claude/code-metrics.yaml (plugins/code-metrics/scripts/resolve-config.py:11-12,250-251) has shipped for more than
  one release (0.6.0, 2026-10-13, plugins/code-metrics/CHANGELOG.md:89,102), so it is dropped, and this repository's copy,
  identical to docs/conventions/code-metrics.yaml, is deleted.
• mutation-testing reads its team layer from docs/conventions/mutation-testing.yaml, validated by a new
  plugins/mutation-testing/schemas/mutation-testing.schema.json, with .claude/mutation-testing.md read only while
  the YAML is absent, for one release. Today .claude/mutation-testing.md is its only team layer
  (plugins/mutation-testing/skills/setup/SKILL.md:26-28), a fenced YAML block in Markdown.

Decisions

• Migrate per-skill .claude config files to userConfig, one conventions doc, or the cascade (ADR 0018) #5906 body, Scope 3: each file moves to its home "with a compatibility read of the old file for one release". The
  2026-10-04 comment sets
  that home to docs/conventions/<concern>.yaml, validated by a JSON Schema.
• ADR 0064 Decision 6: the schema sits at plugins/<plugin>/schemas/<concern>.schema.json, which
  scripts/check-convention-yaml.sh finds by that path (scripts/check-convention-yaml.sh:15-20).
• mutation-testing keeps its key names (tool, command, diff-target, mutate, operators, timeout-ms,
  max-mutants and the rest of the setup template), so a carried value does not change. setup apply writes the YAML
  and, on its first write, carries every key of an existing .claude/mutation-testing.md, as code-metrics' setup does
  (plugins/code-metrics/skills/setup/SKILL.md:84-87).
• This repository's .claude/mutation-testing.md stays until chore(ci): read the post-merge mutation lane's config from docs/conventions/mutation-testing.yaml #12289, because the post-merge mutation lane's test
  reads it (.github/scripts/post-merge-test-mutation/report.test.mjs:304,313), and moving it needs a workflow edit
  the issue lane cannot push (chore(ci): read the post-merge mutation lane's config from docs/conventions/mutation-testing.yaml #12289).
• A GitHub code search finds neither old file in another melodic-software repository, only this one and its eval fixtures.

Key interfaces

• code-metrics: scripts/resolve-config.py and scripts/test_resolve_config.py:463-520, scripts/dispatch.sh:30-31
  and scripts/dispatch.test.sh (its .claude/code-metrics.yaml config cases, for example lines 578 and 687),
  skills/setup/scripts/setup-check.sh:108-111 and setup-check.test.sh:82-85, skills/setup/SKILL.md:14-17,44,
  README.md:77-79, reference/config.md:20-21.
• mutation-testing: skills/setup/SKILL.md (the config merge model, lines 26-28 and 236),
  skills/setup/templates/config-template.md, skills/audit/SKILL.md:191 and skills/audit/context/config-fallback.md,
  which resolve the config through that merge model, README.md:57, and the eval fixture
  skills/audit/evals/fixtures/exercised-scope/.claude/mutation-testing.md with skills/audit/evals/evals.json.
• testing: the cleanup eval fixtures plugins/testing/skills/cleanup/evals/fixtures/cleanup/.claude/mutation-testing.md,
  fixtures/mutation-testing-bash.md, fixtures/mutation-testing-dotnet.md and the prompts in
  plugins/testing/skills/cleanup/evals/evals.json that copy them to .claude/mutation-testing.md.
• Docs: docs/contracts/config-cascade/README.md:650,787, docs/contracts/shell-test-helpers/README.md:122,
  plugins/review/skills/quality-gate/evals/evals.json:238.

Done when

• scripts/run-test-file.sh plugins/code-metrics/scripts/test_resolve_config.py exits 0 with a case where only .claude/code-metrics.yaml exists and the team layer resolves as absent.
• scripts/run-test-file.sh plugins/code-metrics/scripts/dispatch.test.sh exits 0.
• scripts/run-test-file.sh plugins/code-metrics/skills/setup/scripts/setup-check.test.sh exits 0.
• scripts/run-test-file.sh plugins/code-metrics/skills/setup/scripts/test_setup_apply.py exits 0.
• git ls-tree --name-only HEAD .claude/ does not list code-metrics.yaml.
• plugins/mutation-testing/schemas/mutation-testing.schema.json defines every key of the setup template, and scripts/check-convention-yaml.sh --check exits 0.
• git grep -n "docs/conventions/mutation-testing.yaml" -- plugins/mutation-testing/skills/setup/SKILL.md prints the config merge model's team-layer line.
• git grep -nE "\.claude/code-metrics\.yaml" -- plugins docs/contracts ":!*CHANGELOG.md" prints only user-global-layer lines (a ~/, $HOME_DIR or fake-home $home path), setup lines that report or move the retired file, and plugins/source-control/skills/babysit-prs/scripts/tests/test_babysit_risk.py:2554.
• git grep -nE "\.claude/mutation-testing\.md" -- plugins docs/contracts ":!*CHANGELOG.md" prints only lines that name it as the older team file read while docs/conventions/mutation-testing.yaml is absent, setup lines that move it, and lines that name ~/.claude/mutation-testing.md as never read.
• The PR body lists each migrated setting with its new docs/conventions/<concern>.yaml home and schema.
• scripts/check-changelog-fragments.sh --check-required origin/main exits 0.

Out of scope

• The post-merge mutation lane's reader, its test, prompt and workflow comment, this repository's
  docs/conventions/mutation-testing.yaml, and deleting .claude/mutation-testing.md (chore(ci): read the post-merge mutation lane's config from docs/conventions/mutation-testing.yaml #12289).
• The arid-node record .claude/mutation-testing-arid.md, a finding-suppression record, not a setting
  (docs/contracts/config-cascade/README.md:787).
• The .claude/<concern>.local.* personal overlays (ADR 0075 fix 8), which the post-merge lane writes
  (.github/workflows/post-merge-test-mutation.yml:188).
• The user-global ~/.claude/code-metrics.yaml layer (ADR 0075 fix 9).

Posted by the issue-triage session. This was generated by AI.