feat(skills): add terminal milestone handoffs

[Mathews-Tom]

Summary

Adds a required, concrete terminal handoff to Armory's adaptive planning skills.

• plan-prompts now generates a literal NEXT STEPS: block after every milestone GO or NO-GO verdict. The block requires the current milestone action, release-preparation disposition, next runnable milestone or a blocking reason, and NO-GO remediation plus an exact retry gate.
• milestone-runner now validates and reports the same ordered handoff from the observed DAG, external merge state, and declared release train.
• NO-GO results can explicitly bypass the currently blocked closure only for an independent milestone using SKIP <current> FOR NOW; RUN <next> — independent of <closure>.
• Both skills continue to prohibit inferred versioning, tags, publication, and other undeclared release actions.
• Adds focused evaluation cases for terminal next-step behavior and regenerates the relevant manifest entries at version 1.3.0.

No .docs files are included.

Skill Evaluator Results

plan-prompts
  D1 Frontmatter Quality: 20/20
  D2 Trigger Coverage: 18/18
  D3 Structural Completeness: 16/20
  D4 Content Depth: 18/22
  D5 Consistency: 12/12
  D6 Compliance: 8/8
  Overall: 92/100 (92%) — PASS

milestone-runner
  D1 Frontmatter Quality: 20/20
  D2 Trigger Coverage: 18/18
  D3 Structural Completeness: 16/20
  D4 Content Depth: 18/22
  D5 Consistency: 12/12
  D6 Compliance: 8/8
  Overall: 92/100 (92%) — PASS

Verification

• uv run pytest tests/test_generate_adapters.py tests/test_generate_manifest.py -q — 23 passed.
• uv run scripts/validate_frontmatter.py — 135 packages passed.
• uv run scripts/validate_references.py — 135 packages passed.
• uv run python scripts/validate_evals.py — 72 skills, 17 agents, 9 hooks, 6 rules, 7 commands, 3 utilities, and 11 presets passed.
• RED/GREEN/REFACTOR pressure scenarios verified the literal NEXT STEPS: requirement for GO, NO-GO, deferred release preparation, completed release trains, and independent-milestone continuation.

Checklist

• SKILL.md files have valid YAML frontmatter with names and descriptions.
• Skill names remain kebab-case and under 64 characters.
• Descriptions retain Use when clauses and trigger phrases.
• No secrets, credentials, or internal URLs are included.
• All SKILL.md references resolve.
• Both package evaluator scores exceed 70%.
• No CRITICAL or HIGH evaluator findings.
• Tested locally with Claude Code; validation and pressure scenarios ran through the current harness instead.