release skill: Phase 2 step 1 is a single 1,700-word unbroken paragraph #623

Finding

core/surface/skills/release/SKILL.md Phase 2 step 1 (line 240 on fix/579-578-576-findings-triage, will drift as the file grows) is a single unbroken paragraph, 10,256 characters / roughly 1,700 words, with no sub-bullets or line breaks. It composes the annotated tag message, asserts dates-match, and covers several distinct sub-steps in prose form.

Originally raised in #579 (blind exercise runs 14/15, deferred from #576) as "Phase-2 step 1 is one unreadable paragraph."

Why this is real but out of scope for a triage PR

The paragraph is dense specifically because it has accreted HIGH/MEDIUM fix rationale from several adversarial review rounds (each sentence traces to a specific finding). .github/scripts/test_release_lib.py carries a _GOVERNANCE_RULES-style dict of pinned literal phrases from this file's prose (mutation-style checks against specific wording). Restructuring the paragraph into sub-bullets risks:

• silently dropping or rewording a pinned phrase and breaking a governance-rule test in a way that looks like a doc nit but is actually a silent contract regression, and
• losing the "this sentence is the fix for finding X" traceability that a bullet-per-topic split would need to preserve deliberately, not accidentally.

This needs a dedicated pass that (a) maps every pinned phrase to its new location before restructuring, and (b) re-derives the governance-rule test assertions against the new structure — not a drive-by paragraph split bundled into a findings-triage PR touching a dozen other files.

Suggested approach

1. Inventory every _GOVERNANCE_RULES-pinned phrase currently living in this paragraph (.github/scripts/test_release_lib.py).
2. Split the paragraph into a numbered sub-list (6a/6b/6c-style, matching Phase 1's own step-6 sub-lettering conventions used elsewhere in this file), preserving every pinned phrase verbatim.
3. Update the governance-rule tests' location assumptions if any are line/context-sensitive.
4. Re-run test_release_lib.py and the full skill-authoring test suite.

Evidence

$ awk '/^## Phase 2/{f=1} f{print length($0), NR} /^## Phase 3/{exit}' core/surface/skills/release/SKILL.md | sort -rn | head -1
10256 240

Cross-ref: #579, #576.