Greptile Summary

This PR fixes webhook 404s by carrying forward httpRoutes from the previous plugin registry when setActivePluginRegistry is called with a new, route-less registry. The four-condition guard in runtime.ts is conservative and well-targeted: it only activates when the previous registry exists, is a different object, has routes, and the incoming registry has none — matching exactly the runtime reload scenario described in the root cause.

Key points:

• The core logic change in runtime.ts is minimal and correct; the [...prev.httpRoutes] shallow copy appropriately isolates the new registry's array from the old one.
• The fix only carries httpRoutes and intentionally leaves other registry fields (tools, hooks, channels, etc.) alone, which is consistent with the described root cause.
• The three test cases cover the intended scenarios, but they all share the same globalThis singleton state without a beforeEach reset, making them order-dependent. This is the main improvement needed in the test file.

Confidence Score: 4/5

• Safe to merge; the logic change is minimal and conservative with no regressions expected.
• The implementation is a small, well-guarded mutation that precisely addresses the stated root cause. The only concern is test isolation — the test suite mutates shared global state without resetting between cases, which could cause fragile/order-dependent failures as the suite grows. No logic bugs were found in the production code.
• src/plugins/runtime.test.ts — needs a beforeEach reset to prevent test-order sensitivity.

This is a comment left during a code review.
Path: src/plugins/runtime.test.ts
Line: 13-47

Comment:
**Tests share mutable global state across cases**

All three test cases operate against the same `globalThis[Symbol.for("openclaw.pluginRegistryState")]` object, because `setActivePluginRegistry` / `getActivePluginRegistry` read and write from that singleton. The tests happen to pass in the current execution order, but any future re-ordering, parallelism, or new test cases inserted between existing ones could produce unexpected failures because the active registry at the start of each `it` block is whatever was left by the previous one.

Add a `beforeEach` (or `afterEach`) that resets the shared state, for example:

```suggestion
import { describe, it, expect, beforeEach } from "vitest";
import { createEmptyPluginRegistry } from "./registry.js";
import type { PluginHttpRouteRegistration } from "./registry.js";
import { setActivePluginRegistry, getActivePluginRegistry } from "./runtime.js";

const makeRoute = (path: string): PluginHttpRouteRegistration => ({
  path,
  handler: () => {},
  auth: "gateway",
  match: "exact",
});

describe("setActivePluginRegistry", () => {
  beforeEach(() => {
    // Reset the global registry state so each test starts from a clean slate
    setActivePluginRegistry(createEmptyPluginRegistry());
  });
```

How can I resolve this? If you propose a fix, please make it concise.

_{Last reviewed commit: cbd5ccc}

Implemented follow-up for Codex P2 (“Restrict route carry-forward to same registry cache key”) in c0e0be617.

What changed

• src/plugins/runtime.ts

  • setActivePluginRegistry now normalizes and compares cache keys (cacheKey ?? null) before carrying httpRoutes forward.
  • Carry-forward is now limited to same-key registry swaps only.

• src/plugins/runtime.test.ts

  • Added edge-case tests for key matching/mismatching:
    • same key => carry-forward
    • different key => no carry-forward
    • omitted/undefined key => carry-forward (normalized null)
    • null vs explicit key => no carry-forward
  • Kept test-state reset hardening via before/after cleanup.

Validation

• pnpm vitest src/plugins/runtime.test.ts passes locally.

I also observed current CI failures in paths outside this PR diff; if they persist after rerun, I can post a per-job non-causality summary.

CI finished, and I checked the failing jobs against this PR diff.

This PR only changes:

• src/plugins/runtime.ts
• src/plugins/runtime.test.ts

Current failures are in unrelated paths, for example:

• checks (node, extensions, pnpm test:extensions):
  • extensions/feishu/src/client.test.ts
  • extensions/feishu/src/bot.test.ts
  • extensions/google/oauth.test.ts
• checks (node, channels, pnpm test:channels):
  • extensions/slack/src/monitor/provider.ts
  • src/line/bot-handlers.test.ts
• checks (node, contracts, pnpm test:contracts):
  • src/plugins/contracts/catalog.contract.test.ts / src/plugins/provider-runtime.test-support.ts
• checks-windows shards:
  • src/plugins/bundle-mcp.test.ts
  • src/plugin-sdk/index.test.ts

So these failures appear non-causal to this PR’s changes.
Could a maintainer please re-run the failing jobs?

CI failures on this PR still appear non-causal to this diff. Could a maintainer please re-run only the failing jobs from run https://github.com/openclaw/openclaw/actions/runs/23195583617 ?

Requested re-run targets:

• checks (node, extensions, pnpm test:extensions)
• checks (node, channels, pnpm test:channels)
• checks (node, contracts, pnpm test:contracts)
• checks-windows (node, test, 1, 6, pnpm test)
• checks-windows (node, test, 3, 6, pnpm test)
• checks-windows (node, test, 5, 6, pnpm test)
• checks-windows (node, test, 6, 6, pnpm test)

I will not add unrelated "main CI test-fix" commits in this PR, per repo policy (r: no-ci-pr).

Before merge, here is the bug-fix gate checklist from AGENTS.md for this PR:

• Symptom evidence: webhook routes disappear after plugin registry replacement (404 behavior captured in PR discussion/review thread).
• Root cause in code: setActivePluginRegistry replaced registry state and could drop/incorrectly carry route state in src/plugins/runtime.ts.
• Fix touches implicated path: src/plugins/runtime.ts updated carry-forward logic (including cache-key guard).
• Regression test: src/plugins/runtime.test.ts covers same-key carry-forward / different-key no-carry / keyless normalization cases; local pnpm vitest src/plugins/runtime.test.ts was reported passing in-thread.

Merge-policy note: this PR intentionally avoids unrelated CI-only test-fix commits and keeps scope to the runtime registry bug path.

Closing this PR because the issue has been resolved upstream.

This was part of the investigation and an alternative fix approach.
Leaving it here for reference. Thanks.