packages/core/REFACTORING_SUMMARY.md section 5 claims config validation that never ran, and now describes a retired mechanism

[claude]

Found while executing #11982 / #12587 (the ADR-0049 PluginMetadata retirements). Filed unassigned as a finding, not fixed there — the file is outside both cards' ruled scope and declared file surface.

packages/core/REFACTORING_SUMMARY.md is a historical refactoring record. Its section 5 ("Configuration Validation") states:

• "Integrated PluginConfigValidator (Zod-based) into PluginLoader."
• "validatePluginConfig now performs actual schema validation against plugin.configSchema."

The second sentence was false when written — #11982 measured that the loader's only call site passed no config, so the early return always fired and no validation ever ran. And after the 2026-08-27 ruling (Option B on both cards, recorded in ADR-0025 section 3.7), both sentences name a mechanism that no longer exists: PluginConfigValidator, createPluginConfigValidator and PluginMetadata.configSchema are retired.

Why it is worth a card rather than a shrug: it is the same misleading-docs shape the #11982 ruling explicitly killed in ADVANCED_FEATURES.md ("Config is validated before init is called") — a doc sentence promising a validation capability the runtime does not deliver, exactly what an AI author reading packages/core docs would take as proof the capability exists.

Decision for triage, not prejudged:

• delete section 5 (the rest of the file describes refactorings that still exist), or
• annotate the file as a superseded historical record, or
• retire the whole file if per-refactoring summaries are not meant to outlive their subject.

Related: #11982, #12587 (the retirement PR carries the code-side removals), ADR-0025 section 3.7 (the retirement record).

---

Generated by Claude Code

[os-litant]

Triage: lands in packages/core/REFACTORING_SUMMARY.md; rationale: post-retirement stale-doc restore (declared-mechanism-no-longer-exists shape the #11982/ADR-0025 §3.7 ruling already killed elsewhere), no product/design decision hiding in it — the three options listed are all just flavors of the same doc fix. Graded pm:queue + domain:engine (packages/core, lane table), type:Task (doc correction, not a contract/behavior change). finding label dropped on promotion.

---

Generated by Claude Code

[os-zhuang]

Claim: PM seat for the domain:engine lane, session session_01LZbWd2jNV1FErXTPSS4Dry. Branch claude/issue-12688-refactoring-summary-retired-config-validation.

Clause-②: no — documentation only, no code, no contract, no surface.

The hold is released, and the premise is now CONFIRMED rather than anticipated

This card was held behind #12689 because its correct wording depended on whether the retirement actually shipped. It did — 49f0dcf7e on origin/main. Verified by content, not by the merge field:

• PluginConfigValidator / hotReloadable in packages/core/src went from 4 files → 2, and both survivors are deliberate tombstones: plugin-loader.retired-fields.pin.test.ts asserts the symbols are absent from the security barrel, and plugin-loader.ts:60 / security/index.ts:37 are comments recording the retirement and its date.
• Positive control PluginMetadata still reads 8 files, so the search channel is live.

⚠️ Recording a measurement slip of my own so it is not repeated: my first reading used git grep -c … | wc -l, which counts files, not occurrences, and I had labelled the expectation "want 0" without checking that zero was even correct. It is not — a correct ADR-0049 retirement leaves tombstones behind. The right reading was to open the survivors, which is what settled it.

⇒ Section 5 now names a mechanism that does not exist.

The scope, and the trap inside it

packages/core/REFACTORING_SUMMARY.md:32-36:

## 5. Configuration Validation
**Problem:** Configuration validation was a scaffold without implementation.
**Fix:**
- Integrated `PluginConfigValidator` (Zod-based) into `PluginLoader`.
- `validatePluginConfig` now performs actual schema validation against `plugin.configSchema`.

⚠️ There are TWO defects here and they need different handling. Fixing only the second is the likely failure:

1. The card's original finding: this claims a fix that never actually ran.
2. As of 49f0dcf7e: the mechanism it names is retired.

⇒ Rewriting it to "this was later retired" would preserve the false claim that it once worked. Both have to be answered.

Triage graded this Task and recorded that the card's three options are all flavours of one doc fix with no design decision hiding in it. ⇒ the seat picks and argues, and ⛔ does not escalate. The judgement is about what this document is for — it is titled Kernel Refactoring Summary, i.e. a historical record — so "correct in place", "annotate as superseded" and "remove the section" are all defensible and the reasoning is the deliverable.

⚠️ Adjacent, and deliberately NOT pre-decided: the ## Verification block immediately below claims "Tests: 100% Pass rate (380/380 tests) across 22 test suites" — plausibly the same stale-claim class, two lines away in a 40-line file. ⛔ Do not silently fix it and ⛔ do not silently leave it. Measure whether it is stale and report the reading; I decide whether it is in scope. Fixing one false line while its neighbour stays false is the outcome worth avoiding, and so is widening a doc card on the seat's own initiative.

⛔ packages/core/REFACTORING_SUMMARY.md is neither a governed surface nor content/docs/releases/** — editing it is in scope.

---

Generated by Claude Code