fix(skill): auto-generate crystallizer summary/steps instead of rejecting

[kiwipaulrob]

Summary

defaultDraftValidator in core/skill/crystallize.ts throws skill.crystallize.invalid: missing summary / missing steps whenever the LLM returns valid JSON that omits those fields. With deepseek-v4-flash this happens routinely, flooding bridge logs with 10+ ERROR skill.crystallize.failed messages per minute and stalling the crystallizer queue (which in turn delays other RPCs like memory.search waiting on the queue to drain).

This PR makes the validator repair such drafts instead of rejecting them:

• Missing summary → auto-generated from the richest available field, in priority order: first step body → first step title → displayTitle → name → static placeholder "skill procedure". Capped at 200 chars. Never throws for a missing/empty summary.
• Missing/empty steps → a single "Execute the fix" step generated from the summary (title + body, capped at 2000 chars). Only throws when nothing at all can be derived.
• Missing name → still rejected (skill.crystallize.invalid: missing name). On the LLM path this is unreachable anyway — normaliseDraft() already substitutes skill_<policy_id> — so the check remains purely a guard for direct validator use.

Change

• core/skill/crystallize.ts — defaultDraftValidator rewritten: strict → lenient repair (uses || not ?? — LLM JSON emits empty strings, and ?? only falls through on null/undefined).
• tests/unit/skill/crystallize.test.ts — the old "rejects drafts that the validator flags as invalid" test used { steps: [], summary: "" }, which is exactly the case this PR now repairs; converted into an end-to-end regression asserting the draft crystallizes with auto-generated summary + steps.
• tests/unit/skill/crystallize-validator.test.ts — new focused suite (8 tests): full-draft passthrough, never-throws-for-missing-summary, summary fallback chain (step body → displayTitle → name), 200-char cap, empty-string (||) semantics, single-step auto-generation, missing-name rejection.

Tests

• npx vitest run tests/unit/skill → 54 passed (10 files)
• npx tsc -p tsconfig.json --noEmit → clean (exit 0)
• The repair logic is the same shape already proven in production (local Hermes CT100 ran this validator for ~2 weeks; crystallizer backlog drained, zero missing summary rejections since).

Related

• Fixes crystallize: defaultDraftValidator throws skill.crystallize.invalid: missing summary when LLM omits summary field #2143 — the tracked issue for this failure mode (filed 2026-07-22).
• PR fix(skill-crystallize): retry with context when LLM returns empty response #2064 (v2.0.23) addressed the separate empty-response mode; this PR covers the valid-JSON-without-summary mode it did not.

Type of change

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)
• Breaking change (fix or feature that would cause existing functionality to not work as expected)

How Has This Been Tested?

• Unit tests (npx vitest run tests/unit/skill — 54 passed)
• Type-check (tsc --noEmit clean)
• Production-proven shape (the lenient fallback chain has run on Hermes CT100 since ~1 Aug 2026; skill.crystallize.failed missing-summary floods dropped to zero)

Checklist

• I have read the CONTRIBUTING guidelines
• My code follows the style guidelines of this project
• I have performed a self-review of my own code
• I have commented my code, particularly in hard-to-understand areas
• My changes generate no new warnings
• I have added tests that prove my fix is effective
• New and existing unit tests pass locally

[Memtensor-AI]

✅ Automated Test Results: PASSED

All tests passed (15/15 executed). memos_local_plugin/unit: 15/15. Duration: 4s

Branch: fix/crystallizer-validator-auto-fix