[rmax-10]

Summary

Adds one portable skill for ADR-0023, claim-driven software verification:
claims → verification obligations → evidence → decision → residual risk.

Contents:

• portable/claim-driven-software-verification/SKILL.md — the skill. It carries the
  evidence ladder, the risk-to-minimum-evidence mapping, affected-scope verification
  and stopping rules, the fail-closed rule for missing, failing, flaky, or
  unreproducible evidence, and an explicit "no coverage target, no test-count
  target" statement.
• catalog/skills.json — one new entry with ADR-0023 provenance and the
  deterministic source hash. Every other entry is unchanged.
• tests/test_adr_0023_skill.py — new gates: catalog entry and provenance, hash
  parity with the skill directory, required headings, paired **Wrong:**/**Right:**
  cases, checklist length, owned domain terms, cross-references, public-safety
  markers.
• README.md — one row in the ADR-to-skill table.

Handoffs are explicit rather than duplicated: test placement and topology go to
ADR-0025, probabilistic evidence goes to ADR-0026.

Evidence

Measured on the branch head:

python3 -m unittest discover -s tests      -> Ran 36 tests, OK
python3 scripts/catalog_check.py check     -> OK catalog skills=23
python3 -m compileall -q scripts tests     -> exit 0
git diff --check                           -> clean

Negative control: renaming one required heading makes tests/test_adr_0023_skill.py
fail with 2 failures, so the gate is not vacuous.

Provenance and scope

• Card: rmax-ai/delegation-queue#223, round r3.
• Base: 7efbffe (main).
• ADR: rmax-ai/architecture#18, decisions/0023-claim-driven-software-verification-for-agentic-development.md.
• Source only. Materialization into the rmax-10 runtime is a separate post-merge step
  through the existing ADR-0004 route. No profiles/ change is included.

Do not self-merge. Review requested.

---

^{Need help on this PR? Tag @codesmith-bot with what you need.}