PR Review Summary

Status: Drop+Supersede

🪜 Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

• Decision: Drop+Supersede

• Rationale: This docs-only patch encodes an unproven client behavior and presents an in-process test/controller path as an MCP-consumer workaround, while Neo's already-supported explicit-union substrate remains unfalsified. That is a better-existing-substrate + stale-ticket-prescription trigger, so iterating the sentence would preserve the wrong solution shape.

• Disposition: ticket-prescription-off

• Source-coordinate falsifiers: the OpenAPI compiler maps oneOf to z.union; Neural Link already carries string/object unions; and the capability matrix limits call_method to operator/admin or fixture use.

• Salvage map: Keep the Kimi projection receipt and the proof that items: {} reaches Neo's tools/list unchanged. Discard the new description sentence. Move the useful evidence into the amended ticket and use it to test an explicit JsonValue union through Kimi.

• Successor landing pad: Amend #15678 around the explicit-union falsifier, then open a replacement PR whose shape is decided by the Kimi projection result.

• Successor map citation: #15678

Peer-Review Opening: Iris — the cross-harness receipt was valuable: it narrowed the defect to Kimi's projection boundary. The final inference went one step too far, though, and this one-line patch would turn that premature inference into the machine-consumed contract.

🧭 Patch-Blind Premise Snapshot

• Inputs Read Before Patch: #15678, the one-file changed-surface list, current dev OpenAPI/compiler/compliance source, the existing Neural Link oneOf precedent, the capability matrix, and the fleet fixture that owns the in-process call.
• Expected Solution Shape: Before documenting an unavoidable client limitation, test the narrowest explicit JsonValue schema that Neo already knows how to emit. If Kimi still collapses that shape, document only the observable caller-facing limitation and route the client defect upstream; do not point generic MCP callers to a repository-internal service path.
• Patch Verdict: Contradicts the expected shape. The diff changes only the description, asserts that clients are “silently stringifying objects,” and names NeuralLink_InstanceService.callMethod as the workaround, without testing the explicit typed-union path.
• Premise Coherence: Conflicts with verify-before-assert: the evidence proves a narrowed Kimi tool listing, not automatic object stringification or the absence of a Neo-side schema repair.

🕸️ Context & Graph Linking

• Target Epic / Issue ID: Resolves #15678
• Related Graph Nodes: #13372 / #13379 (working OpenAPI union precedent), D#15673 (originating challenge), #15681 (separate write-lock lane)

🔬 Depth Floor

Challenge: I ran the exact Neo conversion path in memory with no tracked-file edit. Replacing only CallMethodRequest.args.items with a union of nullable string, number, boolean, open object, and array emitted:

args.items.anyOf = [string, number, boolean, object(additionalProperties:true), array]

The resulting schema parsed literal strings, numbers, booleans, null, the native {url, bearerToken} object, and nested arrays. That does not prove Kimi will preserve the union, but it disproves “No Neo server code fix exists” as a settled premise. The remaining decisive falsifier is the Kimi tool projection of that explicit schema.

Rhetorical-Drift Audit (per guide §7.4):

• PR description: “No Neo server code fix exists” is not established; the explicit union was not tested.
• OpenAPI description: “silently stringifying objects” overstates the receipt. The evidence shows a string-narrowed agent-facing schema and resulting string calls, not that the client accepts a native object and silently rewrites it.
• Linked workaround: the cited fleet path calls an in-process app handle, which a generic MCP consumer cannot invoke through this tool description.
• No inflated [RETROSPECTIVE] tag or borrowed ticket authority appears in the diff.

Findings: Three substantive framing drifts; all follow from the same wrong solution shape.

🧠 Graph Ingestion Notes

• [KB_GAP]: None.
• [TOOLING_GAP]: The existing strict-client regression proves z.array(z.unknown()) emits an items member; it does not prove a Kimi client preserves an empty schema. A replacement needs an explicit-union compiler assertion plus the Kimi projection receipt.
• [RETROSPECTIVE]: items: {} is permissive server-side but under-specified for clients that insist on a concrete item type. An explicit union is the source-compatible interoperability probe before accepting docs-only degradation.

N/A Audits — 🛂 📜 🔌 🧠

N/A across listed dimensions: this one-line description change introduces no major abstraction, authority-derived demand, wire envelope change, or turn-loaded substrate.

🎯 Close-Target Audit

• Close-target identified: #15678
• #15678 is not epic-labeled
• Delivery is not complete: the interoperability defect remains, the viable explicit-union path is untested, and the upstream client report is optional/unfiled.

Findings: Fails on truthful resolution despite passing the label check. Per close-target remediation, do not downgrade to Refs; supersede this PR with a replacement that can honestly resolve the amended leaf.

📑 Contract Completeness Audit

• The originating ticket has no Contract Ledger separating server emission, Kimi projection, generic MCP access, and in-process fixture access.
• The diff does not match the actual consumer boundary: its workaround is not reachable from the affected MCP caller.

Findings: Contract boundary drift is substantive here because this description is emitted to tool consumers.

🪜 Evidence Audit

• The PR declares Evidence: L1.
• L1 is insufficient for the close-target's runtime cross-harness claim; the decisive explicit-schema projection through Kimi has not been run.
• Residual: none is inaccurate while the interoperability defect and upstream repair remain unresolved.
• The optional post-merge upstream report cannot make the current close-target truthful.

Findings: Evidence/AC mismatch. Green static/unit checks establish YAML validity, not the claimed client behavior or impossibility of a server-side interoperability improvement.

📡 MCP-Tool-Description Budget Audit

• Single-line form retained.
• The description contains an internal class cross-reference that is not a callable MCP path.
• It carries root-cause narrative (“some MCP clients… silently stringifying”) rather than stable call-site usage.
• The 1024-character hard cap is respected.

Findings: Fails audience and narrative discipline even though it fits the byte cap.

🔗 Cross-Skill Integration Audit

• No new MCP tool or startup/skill trigger is introduced.
• The description conflicts with the existing capability boundary: call_method is operator/admin-only and the in-process helper is a trusted fixture/controller surface, not a fallback exposed to an affected MCP client.

Findings: The workaround belongs in controller/fixture documentation if it needs documenting; it does not belong in the generic request-property description.

🧪 Test-Evidence & Location Audit

• Execution evidence: exact-head CI is green at e38bc8d2c4525fb3c29766da6250b78e8ca07d3b; the author's 407-test receipt is consistent with a description-only diff.
• Reviewer falsifier: an in-memory buildZodSchema → toOpenApiJsonSchema probe emitted an explicit object-capable anyOf and parsed all representative JSON value classes.
• The PR has no test for its load-bearing assertion that an explicit schema cannot repair Kimi's projection.
• Test location: N/A for this docs-only diff; the replacement's compiler regression belongs beside OpenApiValidatorCompliance.spec.mjs.

Findings: CI is green but the premise falsifier failed.

📋 Required Actions

To proceed with merging, please address the following:

• Close this PR unmerged and supersede it with a #15678 replacement shaped by the explicit-JsonValue-union Kimi projection: if Kimi preserves the object branch, land the schema plus compliance test; if it still collapses, preserve the raw projection, file the upstream defect, and keep any Neo documentation limited to truthful caller-visible behavior without the in-process-service workaround.

📊 Evaluation Metrics

Verdict weights: 30% premise / right thing, 30% architecture + placement, 30% diff correctness, 10% AC/audit sanity. These are importance-to-verdict weights, not effort budgets.

• [ARCH_ALIGNMENT]: 28 - The change is on a consumed schema surface but points callers across an unavailable boundary and skips the supported explicit-union seam.
• [CONTENT_COMPLETENESS]: 30 - The line covers the observed symptom but overstates cause, workaround accessibility, and repair impossibility.
• [EXECUTION_QUALITY]: 45 - Exact-head CI is green and the edit is syntactically clean; its load-bearing behavioral claims are not tested.
• [PRODUCTIVITY]: 15 - Merging would close the bug without repairing interoperability or establishing an upstream repair path.
• [IMPACT]: 55 - Small diff, but the text is machine-consumed guidance at an admin tool boundary and would steer every affected caller.
• [COMPLEXITY]: 20 - The edit is one line; the real complexity is cross-client schema projection, which the patch avoids.
• [EFFORT_PROFILE]: Maintenance - A bounded schema-compatibility repair, not a documentation-only quick win.

The Kimi receipt found the right boundary. The next move is to make that boundary prove whether an explicit contract survives—not to declare the repair space empty before the probe.

[review-budget-managed]

• outcome: terminal-drop-supersede
• ordinary-limit: 2
• activation-issue: 15257
• activation-pr: 15307
• activated-at: 2026-07-16T20:54:31Z