diff --git a/docs/openclaw-qa-architecture.md b/docs/openclaw-qa-architecture.md
new file mode 100644
index 0000000..63bec86
--- /dev/null
+++ b/docs/openclaw-qa-architecture.md
@@ -0,0 +1,180 @@
+---
+title: OpenClaw QA Harness Architecture
+summary: How the four `@paleo/openclaw-*` packages fit together — bus, gateway, runner, channel plugins, mocked CLIs, artifact layout, and the OpenClaw quirks the harness papers over.
+read_when:
+  - onboarding to the QA-runner codebase
+  - debugging a scenario that misbehaves at the harness layer
+  - touching the Compose stack, the Dockerfile pair, or the mocked-CLI shim
+  - extending a channel plugin or adding a new one
+---
+
+# OpenClaw QA Harness Architecture
+
+Four packages drive automated regression tests against an OpenClaw workspace. Consumers depend on all four; only `openclaw-qa-runner` is the entry point.
+
+| Package | Role |
+| --- | --- |
+| `@paleo/openclaw-qa-runner` | Bus, scenario driver, judge, Compose stack, two-Dockerfile pair, CLI (`init` / `env` / `qa`). |
+| `@paleo/openclaw-channel-mock-core` | Shared channel library — bus client, action handlers, plugin/setup factories, account helpers. Not consumed directly. |
+| `@paleo/openclaw-discord-mock` | Thin wrapper. Registers as channel `discord-mock`, `surface: "discord"`, `autoThread: false`. |
+| `@paleo/openclaw-slack-mock` | Thin wrapper. Registers as channel `slack-mock`, `surface: "slack"`, `autoThread: true`. |
+
+The two wrappers exist side-by-side in one gateway and share a single bus. The runner picks which channel(s) to drive per scenario; `accountId = channelId` keeps per-channel bus state segregated.
+
+## Service topology
+
+Three Compose services. All three are built from the **same image** — only the `command` differs.
+
+```
+            ┌─────────┐
+inbound ──▶ │   bus   │ ◀── outbound (every channel plugin)
+            └────┬────┘
+                 │ HTTP :43123
+       ┌─────────┴─────────┐
+       │                   │
+   ┌───▼───┐           ┌───▼────┐
+   │gateway│           │ runner │
+   └───┬───┘           └───┬────┘
+       │ exec()            │ POST :43124 /mock-cli/invoke
+       ▼                   ▲
+   /opt/qa-mocks/bin ──────┘   (gateway-side shim)
+```
+
+- **`bus`** — in-memory state store. Conversations, threads, messages, events, cursors. Exposes a small HTTP API consumed by `bus-client.ts` in `channel-mock-core`.
+- **`gateway`** — runs `npx openclaw gateway run`. Loads both channel plugins via `plugins.load.paths`. Talks to the bus through its channel plugins; talks to the runner through the mocked-CLI shim.
+- **`runner`** — runs scenarios serially. Mints a fresh `conversationId` per task, pushes inbounds onto the bus, polls outbounds, asserts, runs the judge (Anthropic-direct), writes artifacts.
+
+Healthchecks gate `gateway` on `bus`, and the one-shot `runner` invocation on `gateway`. `runner` is started with `docker compose run --rm --use-aliases runner`; without `--use-aliases` the one-shot container has no network alias and the gateway-side shim's `POST http://runner:43124` fails with `getaddrinfo EAI_AGAIN runner`.
+
+## Two-Dockerfile pattern
+
+`openclaw-qa-runner` ships `Dockerfile.base` (consumer-agnostic): Node 24 Alpine, `claw` user with host-matched UID/GID, mock-CLI shim at `/opt/qa-mocks/`, `/etc/profile` rewritten to keep `/opt/qa-mocks/bin` first in PATH.
+
+The CLI's `env build` builds the base locally as `paleo/openclaw-qa-runner-base:<pkg-version>` and injects the tag into the consumer image via the `QA_RUNNER_BASE_TAG` build arg.
+
+The consumer-owned `Dockerfile` (dropped by `init`) does:
+
+1. `FROM paleo/openclaw-qa-runner-base:${QA_RUNNER_BASE_TAG}`
+2. `COPY` the consumer's `package.json` + `package-lock.json` and `openclaw.json` into the image.
+3. `npm ci --include=dev` — pulls the four `@paleo/openclaw-*` packages from the registry.
+4. `npx openclaw plugins registry --refresh` so the gateway sees the loaded channels.
+5. Optional consumer customizations (extra system packages, skills install, etc.).
+
+`bin/qa` does **not** rebuild. Re-run `npm run env:build` after edits to `openclaw.json` or the consumer `Dockerfile`, or after bumping any `@paleo/openclaw-*` dependency.
+
+`Dockerfile.base` overrides `/etc/profile`. OpenClaw's `exec` tool spawns `/bin/sh -lc <command>`, which sources `/etc/profile`. Alpine's stock profile resets PATH to a "safe" default that drops `/opt/qa-mocks/bin`, silently bypassing the shim — so only commands missing from the default PATH (e.g. `git`, not installed in Alpine) would end up shimmed. Overriding the profile keeps the shim first for every command.
+
+## Compose include
+
+The consumer ships a thin overlay that pulls in the package's base stack:
+
+```yaml
+include:
+  - ./node_modules/@paleo/openclaw-qa-runner/docker-compose.yml
+```
+
+Compose v2.20+ required. The overlay's job is to add consumer-specific service overrides (e.g. extra env vars on `runner`); the base file owns the build context, volumes, healthchecks, and entrypoints.
+
+Path-shaped vars from `.env.local` (`OPENCLAW_WORKSPACE_DIR`, `OPENCLAW_CONFIG_PATH`, `QA_PROJECTS_DIR`, `QA_SCENARIOS_DIR`, `QA_ARTIFACTS_DIR`, `QA_GATEWAY_LOGS_DIR`) are resolved by the CLI against the consumer's `cwd` before invoking Compose — otherwise Compose `include:` would resolve them relative to the package's compose file under `node_modules/`, breaking natural relative paths.
+
+The CLI injects `QA_PROJECT_DIR`, `QA_RUNNER_PACKAGE_DIR`, `CLAW_UID`, `CLAW_GID` automatically.
+
+## Mocked-CLI shim
+
+The gateway's PATH is prepended at runtime with `/opt/qa-mocks/bin/`, where symlinks `git`, `npm`, `pnpm`, `yarn`, `claude` all point at one Node shim. The shim POSTs to `http://runner:43124/mock-cli/invoke` with `{ cli, argv, cwd, stdin }` and replays the JSON response (`{ stdout, stderr, exitCode }`).
+
+The sh wrapper at `/opt/qa-mocks/bin/mock-cli-shim` invokes the shim as `node mock-cli-shim.js "$0" "$@"`. The JS reads the symlink name from `argv[2]` (`/opt/qa-mocks/bin/git` → `git`). Without `"$0"`, the shim would see only the script path and reject every call as `unexpected call to mock-cli-shim.js`.
+
+PATH prepend happens only at gateway runtime — the image build's own `npm install` still uses real `npm`.
+
+Scenarios register handlers via `ctx.mockCli(name, handler)`. Return value: number → exit code; `void`/`undefined` → 0; throw → exit 1 with `handlerError` recorded. Re-registering the same name in one scenario throws. Any invocation with no matching handler **fails the scenario** with `failure.source = "cliMock"` and `message = "unexpected call to <cli>"`, even if no assertion ever ran after.
+
+The runner binds a single in-flight `ConversationRegistry` per scenario; scenarios run serially through one gateway. Each invocation emits a `cliMock` `ReportEvent` carrying the full `CliMockCall` (argv, cwd, stdin, stdout, stderr, exitCode, durationMs, optional handlerError).
+
+## Per-scenario isolation
+
+The bus accumulates state across runs. The only isolation between tasks is the `conversationId` — minted fresh per task as `${scenarioId}-${channel}-${shortRand}` and exposed as `ctx.conversationId`. Scenarios must use `ctx.conversationId` everywhere they currently hard-code a value; metadata that needs to identify the project (e.g. a workspace playbook keying off project name) belongs in the inbound *text*, not in the conversation id.
+
+Scenarios run serially — the base stack ships one `gateway` container; the mocked-CLI shim and runner-side registry are single in-flight.
+
+## Channel plugin internals
+
+Each wrapper exposes two entries in its `package.json`:
+
+- `openclaw.extensions` → `dist/index.js` — the runtime channel plugin (`defineBundledChannelEntry`).
+- `openclaw.setupEntry` → `dist/setup-entry.js` — a setup-only plugin (`defineBundledChannelSetupEntry`).
+
+Both are required. Without `openclaw.setupEntry`, the loader registers the plugin but `resolvePluginRegistrationPlan` skips the `setup-runtime` mode and the channel pipeline never calls `gateway.startAccount`. The setup plugin is a subset of the runtime plugin (`id`, `meta`, `capabilities`, `reload`, `configSchema`, `setup`, `config` — no `messaging` / `gateway` / `actions` / `message`); the loader's `mergeSetupRuntimeChannelPlugin` fills the rest at runtime.
+
+Discovery is wired through `plugins.load.paths` in `openclaw.json`, pointing at the package directory inside the image (`/opt/qa-src/node_modules/@paleo/openclaw-{discord,slack}-mock`). Both plugins must be statically enabled via `plugins.entries["<id>"].enabled = true` — auto-enable for non-bundled (`origin: "config"`) plugins is timing-sensitive against `canStartConfiguredChannelPlugin`: the auto-enable mutation can fire after plan resolution checks `explicitlyEnabled`. Static `enabled: true` makes the check deterministic.
+
+Both channels register together on every gateway boot. The runner selects which to drive per scenario.
+
+`createChannelMockPlugin` in `channel-mock-core` takes `{ channelId, label, surface, autoThread, getRuntime }`. The two wrappers are ten-line modules that bind these knobs:
+
+- `discord-mock` — `surface: "discord"`, `autoThread: false`. Full Discord-shaped surface (`send`, `thread-create`, `thread-reply`, `react`, `read`, `edit`, `delete`, `search`). `thread-create` posts an optional `text`/`message`/`content` atomically with the new thread. Free-form agent text without a tool call lands in the parent channel.
+- `slack-mock` — `surface: "slack"`, `autoThread: true`. Restricted surface (`react` / `read` / `edit` / `delete` / `reactions` / `search`). Bare-channel inbounds auto-thread on the triggering message; every subsequent outbound from the same turn lands in that thread.
+
+Inbound metadata claims `Provider` / `Surface` / `OriginatingChannel` = the registered channel id, so the SDK routes tool-schema discovery back to the right plugin. `chat_id` envelope shape is **not** rewritten — scenarios assert on `conversation.id` / `threadId`, not envelope formatting.
+
+`openclaw.plugin.json` without `channelConfigs` warns at startup (`channel plugin manifest declares <id> without channelConfigs metadata`); the gateway fills missing `label` / `selectionLabel` / `docsPath` / `blurb` from the runtime plugin. Cosmetic.
+
+## Target normalizer + plugin-action vs send
+
+OpenClaw's `normalizeMessageActionInput` runs before any `"to"`-mode plugin handler (`send`, `thread-create`, `thread-reply`, `react`, `read`, `edit`, `delete`). It rewrites `channelId` → `target` → `to` and deletes the original `channelId` key. A handler that reads `channelId` directly is broken-by-construction. `channel-mock-core`'s `resolveDestination` always reads `to` first.
+
+Canonical destination param is `to`. Accepted shapes:
+
+- `channel:<id>` or bare `<id>` (channel)
+- `dm:<id>`
+- `group:<id>`
+- `thread:<channelId>/<threadId>`
+
+Resolved in the order `to → target → channelId` to match the normalizer's output.
+
+Plugin actions and `send` route through different handlers in `message-action-runner.ts`. Only `send` triggers the delivery mirror, which historically tripped a lock-fence race (`EmbeddedAttemptSessionTakeoverError`). Plugin actions don't set `ctx.mirror` and never trip the race. Workspace-driven outbound that needs a thread should use `thread-create` + `thread-reply` rather than `send`.
+
+`BindingMatchSchema` is strict-equality on `peer.id`. No catch-all binding without multi-account channel config. The judge agent (in OpenClaw config) is left config-only and never instantiated; the actual judge runs out-of-process from the runner against Anthropic directly.
+
+## Artifacts & cost
+
+Layout: `artifacts/<runStamp>/<scenario>-<channel>[-<NN>][-<VERDICT>]/`.
+
+- `<NN>` — iteration index, padded to the width of `--iterations`. Omitted when `--iterations 1`.
+- `<VERDICT>` — `PASS` / `FAIL`. Applied by **renaming the directory** after `report.json` lands. A directory with no verdict suffix means the run is pending or crashed before the rename.
+
+Two files per task:
+
+- `events.jsonl` — appended live as the scenario runs, one `ReportEvent` per line. Survives a runner crash. Original write-order `seq`s preserved.
+- `report.json` — final `ScenarioReport`, written once at end. Re-merges the live events with `agentToolCall` entries parsed from the gateway's `anthropic-payload.jsonl` (filtered by `conversationId`), re-assigns `seq` by `ts`, and adds per-scenario `cost = { gatewayUsd, judgeUsd, totalUsd, gatewayTurns }`.
+
+Event kinds: `log` · `inboundSent` · `outboundReceived` · `assertion` · `judge` · `cliMock` · `agentToolCall` · `failure`. `outboundReceived` captures every bus outbound for the conversation, not only the ones the scenario explicitly awaits. `agentToolCall` lives only in `report.json`.
+
+Authoritative types: `packages/openclaw-qa-runner/src/report.ts`.
+
+Cost: the runner sums the gateway's `stage:"usage"` entries from `anthropic-payload.jsonl` for any entry with `ts >= runStart`, plus the judge's inline `usage` priced via an in-runner table. A 5s grace wait after the last task lets OpenClaw flush its usage record (it lands ~2s after the outbound hits the bus). Failing runs that time out before the agent completes report `$0.0000` — the gateway never wrote a usage record for the unfinished turn.
+
+`OPENCLAW_ANTHROPIC_PAYLOAD_LOG=1` is forced on by the Compose stack — QA needs the file. `OPENCLAW_RAW_STREAM=1` is opt-in for `raw-stream.jsonl`. Both land under `.gateway-logs/` (bind-mounted from `~/.openclaw/logs/`).
+
+## Judge
+
+`judgeLLM` calls Anthropic directly from the runner — no bus traffic, no gateway involvement. Not an OpenClaw agent. Model defaults to `anthropic/claude-haiku-4-5`; override via `QA_JUDGE_MODEL` on the runner service. LiteLLM-style ref required; only the `anthropic/` provider is wired up today.
+
+Prefer structural assertions over `judgeLLM`; reserve the judge for free-form content claims.
+
+## OpenClaw config quirks the harness depends on
+
+- **`agents.list[*].workspace`, not `workspaceDir`.** The schema accepts both spellings on related surfaces, but the agents list only reads `workspace`.
+- **`gateway.mode: "local"` required.** Without it, startup fails with `existing config is missing gateway.mode`.
+
+## Scenario loading
+
+Scenarios are `.ts` files under `scenarios/`, default-export `async (ctx: ScenarioContext) => void`. Loaded at runtime by Node 24's built-in TypeScript stripping (the image uses Node 24). Stick to the strip-compatible subset: type annotations, `as`, `satisfies`, generics, interfaces. Avoid `enum`, `namespace`, constructor parameter properties, decorators, `import =`.
+
+`discoverScenarios()` filters on `.ts` suffix on file entries only — directories under `scenarios/` (e.g. `_lib/`) are ignored, which is the idiomatic place for shared scenario helpers.
+
+## See also
+
+- Each package's `README.md` — actionable usage.
+- `packages/openclaw-qa-runner/src/context.ts` — `ScenarioContext` definition.
+- `packages/openclaw-qa-runner/src/report.ts` — authoritative event/report types.
diff --git a/package-lock.json b/package-lock.json
index 29514ef..98fcf78 100644
--- a/package-lock.json
+++ b/package-lock.json
@@ -8480,7 +8480,7 @@
     },
     "packages/docmap": {
       "name": "@paleo/docmap",
-      "version": "0.4.2",
+      "version": "0.4.3",
       "license": "CC0-1.0",
       "bin": {
         "docmap": "bin/docmap.mjs"
@@ -8497,7 +8497,7 @@
     },
     "packages/openclaw-channel-mock-core": {
       "name": "@paleo/openclaw-channel-mock-core",
-      "version": "0.2.2",
+      "version": "0.2.3",
       "license": "MIT",
       "dependencies": {
         "typebox": "1.1.38"
@@ -8518,10 +8518,10 @@
     },
     "packages/openclaw-discord-mock": {
       "name": "@paleo/openclaw-discord-mock",
-      "version": "0.2.2",
+      "version": "0.2.3",
       "license": "MIT",
       "dependencies": {
-        "@paleo/openclaw-channel-mock-core": "0.2.2"
+        "@paleo/openclaw-channel-mock-core": "0.2.3"
       },
       "devDependencies": {
         "@types/node": "~24.12.4",
@@ -8539,13 +8539,13 @@
     },
     "packages/openclaw-qa-runner": {
       "name": "@paleo/openclaw-qa-runner",
-      "version": "0.4.0",
+      "version": "0.4.1",
       "license": "MIT",
       "dependencies": {
         "@anthropic-ai/sdk": "~0.97.1",
-        "@paleo/openclaw-channel-mock-core": "0.2.2",
-        "@paleo/openclaw-discord-mock": "0.2.2",
-        "@paleo/openclaw-slack-mock": "0.2.2"
+        "@paleo/openclaw-channel-mock-core": "0.2.3",
+        "@paleo/openclaw-discord-mock": "0.2.3",
+        "@paleo/openclaw-slack-mock": "0.2.3"
       },
       "bin": {
         "openclaw-qa-runner": "bin/cli.mjs"
@@ -8565,10 +8565,10 @@
     },
     "packages/openclaw-slack-mock": {
       "name": "@paleo/openclaw-slack-mock",
-      "version": "0.2.2",
+      "version": "0.2.3",
       "license": "MIT",
       "dependencies": {
-        "@paleo/openclaw-channel-mock-core": "0.2.2"
+        "@paleo/openclaw-channel-mock-core": "0.2.3"
       },
       "devDependencies": {
         "@types/node": "~24.12.4",
diff --git a/packages/docmap/CHANGELOG.md b/packages/docmap/CHANGELOG.md
index ac1fa6d..d235917 100644
--- a/packages/docmap/CHANGELOG.md
+++ b/packages/docmap/CHANGELOG.md
@@ -1,5 +1,11 @@
 # @paleo/docmap
 
+## 0.4.3
+
+### Patch Changes
+
+- Improved documentation
+
 ## 0.4.2
 
 ### Patch Changes
diff --git a/packages/docmap/README.md b/packages/docmap/README.md
index ca30a3f..0256e88 100644
--- a/packages/docmap/README.md
+++ b/packages/docmap/README.md
@@ -84,4 +84,4 @@ npx @paleo/docmap --root path/to/docs
 | `--check` | Validate all files and directories. Reports name and frontmatter issues. |
 | `--root <path>` | Use a custom directory as the docs root instead of `docs/`. |
 
-For internals, see [docs/docmap-architecture.md](../../docs/docmap-architecture.md).
+For internals, see [docs/docmap-architecture.md](https://github.com/paleo/alignfirst/blob/main/docs/docmap-architecture.md).
diff --git a/packages/docmap/package.json b/packages/docmap/package.json
index ce4fe6a..cd550fc 100644
--- a/packages/docmap/package.json
+++ b/packages/docmap/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@paleo/docmap",
-  "version": "0.4.2",
+  "version": "0.4.3",
   "license": "CC0-1.0",
   "description": "A lightweight documentation system for AI agents and humans.",
   "keywords": [
diff --git a/packages/openclaw-channel-mock-core/CHANGELOG.md b/packages/openclaw-channel-mock-core/CHANGELOG.md
index fe97930..596eedb 100644
--- a/packages/openclaw-channel-mock-core/CHANGELOG.md
+++ b/packages/openclaw-channel-mock-core/CHANGELOG.md
@@ -1,5 +1,11 @@
 # @paleo/openclaw-channel-mock-core
 
+## 0.2.3
+
+### Patch Changes
+
+- Improved documentation
+
 ## 0.2.2
 
 ### Patch Changes
diff --git a/packages/openclaw-channel-mock-core/README.md b/packages/openclaw-channel-mock-core/README.md
index 414c660..bc8a51a 100644
--- a/packages/openclaw-channel-mock-core/README.md
+++ b/packages/openclaw-channel-mock-core/README.md
@@ -4,10 +4,10 @@ Shared library powering the synthetic OpenClaw channel plugins used in QA harnes
 
 Not meant to be consumed directly. Use the surface wrappers:
 
-- [`@paleo/openclaw-discord-mock`](../openclaw-discord-mock/) — `surface: "discord"`, full action surface, `autoThread: false`.
-- [`@paleo/openclaw-slack-mock`](../openclaw-slack-mock/) — `surface: "slack"`, restricted action surface, `autoThread: true`.
+- [`@paleo/openclaw-discord-mock`](https://www.npmjs.com/package/@paleo/openclaw-discord-mock) — `surface: "discord"`, full action surface, `autoThread: false`.
+- [`@paleo/openclaw-slack-mock`](https://www.npmjs.com/package/@paleo/openclaw-slack-mock) — `surface: "slack"`, restricted action surface, `autoThread: true`.
 
-Both wrappers register as OpenClaw channels and talk to a single bus (`http://bus:43123` by default) provisioned by [`@paleo/openclaw-qa-runner`](../openclaw-qa-runner/).
+Both wrappers register as OpenClaw channels and talk to a single bus (`http://bus:43123` by default) provisioned by [`@paleo/openclaw-qa-runner`](https://www.npmjs.com/package/@paleo/openclaw-qa-runner).
 
 ## Attribution
 
diff --git a/packages/openclaw-channel-mock-core/package.json b/packages/openclaw-channel-mock-core/package.json
index 06dfe02..31e2dfe 100644
--- a/packages/openclaw-channel-mock-core/package.json
+++ b/packages/openclaw-channel-mock-core/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@paleo/openclaw-channel-mock-core",
-  "version": "0.2.2",
+  "version": "0.2.3",
   "description": "Shared library for synthetic OpenClaw channel plugins used in QA harnesses (bus client, action handlers, factories).",
   "keywords": [
     "openclaw",
diff --git a/packages/openclaw-discord-mock/CHANGELOG.md b/packages/openclaw-discord-mock/CHANGELOG.md
index b4253f4..0828970 100644
--- a/packages/openclaw-discord-mock/CHANGELOG.md
+++ b/packages/openclaw-discord-mock/CHANGELOG.md
@@ -1,5 +1,13 @@
 # @paleo/openclaw-discord-mock
 
+## 0.2.3
+
+### Patch Changes
+
+- Improved documentation
+- Updated dependencies
+  - @paleo/openclaw-channel-mock-core@0.2.3
+
 ## 0.2.2
 
 ### Patch Changes
diff --git a/packages/openclaw-discord-mock/README.md b/packages/openclaw-discord-mock/README.md
index ff1dec2..aa7f017 100644
--- a/packages/openclaw-discord-mock/README.md
+++ b/packages/openclaw-discord-mock/README.md
@@ -1,10 +1,45 @@
 # @paleo/openclaw-discord-mock
 
-Synthetic Discord-shaped OpenClaw channel plugin. Registers as channel `discord-mock` with the full Discord action surface (`send`, `thread-create`, `thread-reply`, `react`, `read`, `edit`, `delete`, `search`). `thread-create` posts an optional body atomically with the new thread; free-form agent text without a tool call lands in the parent channel.
+Synthetic Discord-shaped OpenClaw channel plugin. Registers as channel `discord-mock`. Full Discord-shaped action surface: `send`, `thread-create`, `thread-reply`, `react`, `read`, `edit`, `delete`, `search`. `thread-create` posts an optional `text` / `message` / `content` atomically with the new thread; free-form agent text without a tool call lands in the parent channel.
 
-Backed by [`@paleo/openclaw-channel-mock-core`](../openclaw-channel-mock-core/) with `surface: "discord"` and `autoThread: false`. Pair with [`@paleo/openclaw-qa-runner`](../openclaw-qa-runner/) for the QA harness.
+Backed by [`@paleo/openclaw-channel-mock-core`](https://www.npmjs.com/package/@paleo/openclaw-channel-mock-core) (`surface: "discord"`, `autoThread: false`). Pair with [`@paleo/openclaw-qa-runner`](https://www.npmjs.com/package/@paleo/openclaw-qa-runner) for the QA harness.
 
-`Provider` / `Surface` / `OriginatingChannel` on inbound metadata are claimed as `discord-mock` so the SDK routes tool-schema discovery to this plugin.
+## Install
+
+```sh
+npm i -D @paleo/openclaw-discord-mock
+```
+
+The runner depends on this package transitively — installing `@paleo/openclaw-qa-runner` already pulls it in.
+
+## Enable
+
+In your `openclaw.json`:
+
+```json
+{
+  "plugins": {
+    "load": { "paths": ["/opt/qa-src/node_modules/@paleo/openclaw-discord-mock"] },
+    "entries": { "discord-mock": { "enabled": true } }
+  },
+  "channels": {
+    "discord-mock": {
+      "baseUrl": "http://bus:43123",
+      "botUserId": "openclaw",
+      "botDisplayName": "OpenClaw QA",
+      "allowFrom": ["*"]
+    }
+  }
+}
+```
+
+`enabled: true` must be **static**. Auto-enable for `origin: "config"` plugins is timing-sensitive against the plan-resolution `explicitlyEnabled` check.
+
+## Target format
+
+Canonical destination is the `to` param. Accepts `channel:<id>` / bare `<id>` / `dm:<id>` / `group:<id>` / `thread:<channelId>/<threadId>`.
+
+`Provider` / `Surface` / `OriginatingChannel` on inbound metadata are claimed as `discord-mock` so the SDK routes tool-schema discovery to this plugin. The `chat_id` envelope shape is **not** rewritten — assert on `conversation.id` / `threadId`.
 
 ## Attribution
 
diff --git a/packages/openclaw-discord-mock/package.json b/packages/openclaw-discord-mock/package.json
index 701304d..7291bd7 100644
--- a/packages/openclaw-discord-mock/package.json
+++ b/packages/openclaw-discord-mock/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@paleo/openclaw-discord-mock",
-  "version": "0.2.2",
+  "version": "0.2.3",
   "description": "Synthetic Discord-shaped OpenClaw channel plugin for automated QA scenarios.",
   "keywords": [
     "openclaw",
@@ -63,7 +63,7 @@
     "openclaw": "*"
   },
   "dependencies": {
-    "@paleo/openclaw-channel-mock-core": "0.2.2"
+    "@paleo/openclaw-channel-mock-core": "0.2.3"
   },
   "devDependencies": {
     "@types/node": "~24.12.4",
diff --git a/packages/openclaw-qa-runner/CHANGELOG.md b/packages/openclaw-qa-runner/CHANGELOG.md
index bcfb678..e9e4d54 100644
--- a/packages/openclaw-qa-runner/CHANGELOG.md
+++ b/packages/openclaw-qa-runner/CHANGELOG.md
@@ -1,5 +1,15 @@
 # @paleo/openclaw-qa-runner
 
+## 0.4.1
+
+### Patch Changes
+
+- Improved documentation
+- Updated dependencies
+  - @paleo/openclaw-channel-mock-core@0.2.3
+  - @paleo/openclaw-discord-mock@0.2.3
+  - @paleo/openclaw-slack-mock@0.2.3
+
 ## 0.4.0
 
 ### Minor Changes
diff --git a/packages/openclaw-qa-runner/README.md b/packages/openclaw-qa-runner/README.md
index 8a5302b..11d4e5d 100644
--- a/packages/openclaw-qa-runner/README.md
+++ b/packages/openclaw-qa-runner/README.md
@@ -1,8 +1,10 @@
 # @paleo/openclaw-qa-runner
 
-Dockerised regression-test harness for OpenClaw workspaces. Drives the agent through two synthetic channels (`discord-mock`, `slack-mock`) and asserts the results. One gateway, one bus, parallel scenarios.
+Dockerised regression-test harness for OpenClaw workspaces. Drives the agent through two synthetic channels (`discord-mock`, `slack-mock`) and asserts the results.
 
-Toolkit entry point. Pair with [`@paleo/openclaw-channel-mock-core`](../openclaw-channel-mock-core/), [`@paleo/openclaw-discord-mock`](../openclaw-discord-mock/), [`@paleo/openclaw-slack-mock`](../openclaw-slack-mock/).
+Pair with [`@paleo/openclaw-channel-mock-core`](https://www.npmjs.com/package/@paleo/openclaw-channel-mock-core), [`@paleo/openclaw-discord-mock`](https://www.npmjs.com/package/@paleo/openclaw-discord-mock), [`@paleo/openclaw-slack-mock`](https://www.npmjs.com/package/@paleo/openclaw-slack-mock).
+
+For internals (topology, Dockerfile pair, mocked-CLI shim, channel plugin mechanics, OpenClaw quirks), see [openclaw-qa-architecture.md](https://github.com/paleo/alignfirst/blob/main/docs/openclaw-qa-architecture.md).
 
 ## Install
 
@@ -10,22 +12,9 @@ Toolkit entry point. Pair with [`@paleo/openclaw-channel-mock-core`](../openclaw
 npm i -D @paleo/openclaw-qa-runner @paleo/openclaw-channel-mock-core @paleo/openclaw-discord-mock @paleo/openclaw-slack-mock openclaw
 ```
 
-Requires Docker Compose v2.20+ (consumer overlay uses Compose `include:`).
-
-## Init
-
-```sh
-npx @paleo/openclaw-qa-runner init .
-```
-
-Drops four files into the target directory:
-
-- `openclaw.json` — gateway config (mode `local`, both channel plugins enabled, main agent).
-- `.env.local.example` — copy to `.env.local`, set `ANTHROPIC_API_KEY` and `OPENCLAW_WORKSPACE_DIR` (host path to your OpenClaw workspace).
-- `docker-compose.yml` — thin overlay that `include:`s this package's base stack from `node_modules/`.
-- `Dockerfile` — consumer-owned image. Inherits the package's base via `FROM paleo/openclaw-qa-runner-base:${QA_RUNNER_BASE_TAG}` (the tag is injected by the CLI from the installed package version). Add `RUN`/`COPY`/`ENV` directives for any consumer-specific setup (skill installs, extra system packages, etc.).
+Requires Docker Compose v2.20+ (overlay uses Compose `include:`).
 
-Then wire `package.json` scripts:
+Wire `package.json` scripts:
 
 ```json
 "scripts": {
@@ -36,85 +25,116 @@ Then wire `package.json` scripts:
 }
 ```
 
-Each command derives `QA_PROJECT_DIR` from `cwd`, `QA_RUNNER_PACKAGE_DIR` from its own install location, and `CLAW_UID`/`CLAW_GID` from the host user — no boilerplate in `package.json`.
+## Init
+
+```sh
+npx @paleo/openclaw-qa-runner init <qa-dir>
+```
+
+Drops four files:
+
+- `openclaw.json` — gateway config (mode `local`, both channel plugins enabled, main agent placeholder).
+- `.env.local.example` — copy to `.env.local`, fill `ANTHROPIC_API_KEY` + `OPENCLAW_WORKSPACE_DIR`.
+- `docker-compose.yml` — thin overlay that `include:`s the base from `node_modules/`.
+- `Dockerfile` — consumer-owned. Inherits the base via `FROM paleo/openclaw-qa-runner-base:${QA_RUNNER_BASE_TAG}`. Add `RUN`/`COPY`/`ENV` for consumer-specific setup (extra system packages, skills install, etc.).
 
 ## Configure
 
 Edit `openclaw.json`:
 
-- `agents.list[id=main].model` — LiteLLM-style `provider/model` ref (e.g. `anthropic/claude-sonnet-4-6`). The template ships a placeholder; OpenClaw will fail loudly until you pick one.
-- `agents.list[id=main].workspace` — host path to your OpenClaw workspace (bind-mounted into the gateway).
+- `agents.list[id=main].model` — LiteLLM-style `provider/model` ref. The template ships a placeholder; OpenClaw fails loudly until you pick one.
+- `agents.list[id=main].workspace` — host path to your OpenClaw workspace, bind-mounted into the gateway. Field name is **`workspace`**, not `workspaceDir`.
 - `channels.*` — both `discord-mock` and `slack-mock` blocks point at the same bus.
 
-The LLM judge runs out-of-process — Anthropic-direct from the runner, never through the gateway — so it is **not** an OpenClaw agent and is not configured via `openclaw.json`. Defaults to `anthropic/claude-haiku-4-5`; override via the `QA_JUDGE_MODEL` env var on the runner. The ref must be LiteLLM-style; only the `anthropic/` provider is wired up today.
+Drop scenarios under `scenarios/<id>.ts`. Project fixtures under `projects-fixture/` (bind-mounted to `~/projects/` in the gateway).
+
+Scenarios are loaded by Node 24's built-in TypeScript stripping. Stick to the strip-compatible subset (no `enum`, `namespace`, decorators, ctor parameter properties, `import =`). Shared helpers go under `scenarios/_lib/` — `discoverScenarios()` skips directories.
+
+## Env vars (`.env.local`)
 
-Drop scenarios under `scenarios/<id>.ts`, default-export `async (ctx: ScenarioContext) => void`. Project fixtures live under `projects-fixture/` (bind-mounted to `~/projects/`).
+Required:
 
-Scenarios are loaded at runtime by Node's built-in TypeScript stripping (Node 24, which the image uses). Stick to the strip-compatible subset: type annotations, `as`, `satisfies`, generics, interfaces. Avoid `enum`, `namespace`, constructor parameter properties, decorators, and `import =`.
+- `ANTHROPIC_API_KEY`
+- `OPENCLAW_WORKSPACE_DIR` — host path mounted at `/home/claw/.openclaw/workspace`.
 
-## Build / up / run
+Optional (defaults relative to the consumer's `qa/` dir):
 
-`env:build` first builds the consumer-agnostic base image (`paleo/openclaw-qa-runner-base:<pkg-version>`, locally tagged) from this package's `Dockerfile.base`, then runs `docker compose build` against the consumer's `Dockerfile`. Docker layer cache makes repeat base builds near-free; `env:up` / `qa` skip the base build when the tag already exists.
+- `OPENCLAW_CONFIG_PATH` → `./openclaw.json`
+- `QA_PROJECTS_DIR` → `./projects-fixture`
+- `QA_SCENARIOS_DIR` → `./scenarios`
+- `QA_ARTIFACTS_DIR` → `./artifacts`
+- `QA_GATEWAY_LOGS_DIR` → `./.gateway-logs`
+- `OPENCLAW_RAW_STREAM=1` — also write `raw-stream.jsonl` alongside the always-on `anthropic-payload.jsonl`.
+
+`QA_PROJECT_DIR`, `QA_RUNNER_PACKAGE_DIR`, `CLAW_UID`, `CLAW_GID` are injected by the CLI.
+
+## Run
 
 ```sh
-npm run env:build                                                  # build the gateway / bus / runner image
+npm run env:build                                                  # build base + consumer image
 npm run env:up                                                     # bring up bus + gateway (both channels register)
 npm run qa -- --channel all <scenario>                             # one scenario, both channels
 npm run qa -- --channel all --all                                  # every scenario, both channels
 npm run qa -- --channel discord-mock <scenario>                    # restrict to one channel
-npm run qa -- --channel all --iterations 5 <scenario>              # repeat each (scenario, channel) pair 5 times
+npm run qa -- --channel all --iterations 5 <scenario>              # repeat each (scenario, channel) pair 5×
 npm run qa -- --channel all --iterations 5 --max-failures 1 <s>    # abort a pair after >1 failure
 npm run env:down
 ```
 
-Artifacts land under `artifacts/<runStamp>/<scenario>-<channel>[-<NN>][-<VERDICT>]/`: `<NN>` is the iteration index (omitted when `--iterations 1`), `<VERDICT>` is `PASS` / `FAIL`, applied by renaming the dir after `report.json` is written — its absence means the run is still pending or crashed before rename. Exit 0 iff every pair passes.
+`env:build` first builds the base image (`paleo/openclaw-qa-runner-base:<pkg-version>`) from this package's `Dockerfile.base`, then builds the consumer image. Layer cache makes repeat base builds near-free; `env:up` / `qa` skip the base build when the tag already exists.
+
+Rebuild required after: bumping any `@paleo/openclaw-*` dependency, edits to `openclaw.json`, or any change to the consumer `Dockerfile`.
+
+Scenarios run **serially** through one gateway. Exit 0 iff every pair passes.
 
 ## Scenario primitives
 
 From `@paleo/openclaw-qa-runner` (`src/context.ts`):
 
 - `channel`, `conversationId`, `accountId` — per-task isolation. Use `ctx.conversationId` everywhere; never hard-code a value.
-- `sendInbound(text, opts?)` — push an inbound on the bus.
-- `poll(opts?)`, `waitForOutbound(opts?)`, `expectNoOutbound(opts?)` — bus consumers.
+- `sendInbound(input)` — push an inbound on the bus.
+- `poll`, `waitForOutbound`, `expectNoOutbound` — bus consumers.
 - `assertRegex`, `assertEqual`, `assertLength` — structural assertions.
-- `judgeLLM(prompt)` — Anthropic-direct judgement (no bus traffic).
+- `judgeLLM({ message, rubric, label })` — Anthropic-direct judgement (no bus traffic, no gateway).
+- `mockCli(name, handler)` — intercepts the gateway's calls to `git` / `npm` / `pnpm` / `yarn` / `claude`. Unregistered calls fail the scenario with `failure.source = "cliMock"`.
 - `log`, `getCursor`.
 
 Prefer structural assertions over `judgeLLM`; reserve the judge for free-form content claims.
 
-## Channels
+## Judge model
 
-Both plugins register together. Pick which to drive per scenario via `--channel discord-mock|slack-mock|all`.
+Defaults to `anthropic/claude-haiku-4-5`. Override via `QA_JUDGE_MODEL` on the `runner` service (set in your consumer overlay). The ref must be LiteLLM-style; only the `anthropic/` provider is wired up today. The judge is **not** an OpenClaw agent — don't configure it in `openclaw.json`.
 
-- `discord-mock` — full Discord-shaped surface; `thread-create` posts an optional body atomically.
-- `slack-mock` — restricted Slack-shaped surface (`react` / `read` / `edit` / `delete` / `reactions` / `search`). Bare-channel inbounds auto-thread on the triggering message.
+## Artifacts
 
-## Compose stack
+`artifacts/<runStamp>/<scenario>-<channel>[-<NN>][-<VERDICT>]/`:
 
-The CLI sets `QA_PROJECT_DIR`, `QA_RUNNER_PACKAGE_DIR`, `CLAW_UID`, `CLAW_GID` automatically. Everything else comes from `.env.local`:
+- `events.jsonl` — appended live, survives a runner crash.
+- `report.json` — final `ScenarioReport`. Merges `events.jsonl` with `agentToolCall` entries from the gateway payload log; adds per-scenario `cost`.
 
-- `ANTHROPIC_API_KEY` — required.
-- `OPENCLAW_WORKSPACE_DIR` — required (host path mounted at `/home/claw/.openclaw/workspace`).
-- `OPENCLAW_CONFIG_PATH` — default `<qa>/openclaw.json` → `/home/claw/.openclaw/openclaw.json`.
-- `QA_PROJECTS_DIR` — default `<qa>/projects-fixture` → `/home/claw/projects/`.
-- `QA_SCENARIOS_DIR` — default `<qa>/scenarios` → `/opt/qa-src/scenarios`.
-- `QA_ARTIFACTS_DIR` — default `<qa>/artifacts` → `/opt/qa-artifacts`.
-- `QA_GATEWAY_LOGS_DIR` — default `<qa>/.gateway-logs` → `/home/claw/.openclaw/logs`.
+`<NN>` is the iteration index (omitted when `--iterations 1`). `<VERDICT>` is `PASS` / `FAIL`, applied by **renaming the directory** after `report.json` is written. A directory with no verdict suffix means the run is pending or crashed before rename.
 
-`<qa>` is the consumer's qa dir (the wrapper's `cwd`). Rebuild this package's `dist/` with `npm run build` to refresh the mount.
+Authoritative types: `src/report.ts`.
 
-Healthchecks: `gateway` waits on `bus`, `runner` waits on `gateway`.
+## Channels
 
-## Gateway logs (opt-in)
+Both `discord-mock` and `slack-mock` register on every boot. Pick which to drive per scenario via `--channel discord-mock|slack-mock|all`.
 
-Set in `.env.local`:
+- `discord-mock` — full Discord-shaped surface; `thread-create` posts an optional body atomically.
+- `slack-mock` — restricted Slack-shaped surface (`react` / `read` / `edit` / `delete` / `reactions` / `search`). Bare-channel inbounds auto-thread on the triggering message.
 
-```sh
-OPENCLAW_ANTHROPIC_PAYLOAD_LOG=1
-OPENCLAW_RAW_STREAM=1
-```
+Inbound metadata claims `Provider` / `Surface` / `OriginatingChannel` = the registered channel id so the SDK routes tool-schema discovery to the right plugin. Assert on `conversation.id` / `threadId`, not envelope formatting.
+
+## Target format
+
+Canonical destination param is `to`. Accepted shapes:
+
+- `channel:<id>` or bare `<id>` (channel)
+- `dm:<id>`
+- `group:<id>`
+- `thread:<channelId>/<threadId>`
 
-Writes `anthropic-payload.jsonl` and `raw-stream.jsonl` under `.gateway-logs/`. The runner's cost reporting reads `anthropic-payload.jsonl`.
+Actions resolve `to → target → channelId`.
 
 ## Attribution
 
diff --git a/packages/openclaw-qa-runner/package.json b/packages/openclaw-qa-runner/package.json
index 5b152a4..1f4744d 100644
--- a/packages/openclaw-qa-runner/package.json
+++ b/packages/openclaw-qa-runner/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@paleo/openclaw-qa-runner",
-  "version": "0.4.0",
+  "version": "0.4.1",
   "description": "Dockerised regression-test harness for OpenClaw workspaces: bus, scenario driver, judge, Compose stack.",
   "keywords": [
     "openclaw",
@@ -47,9 +47,9 @@
   },
   "dependencies": {
     "@anthropic-ai/sdk": "~0.97.1",
-    "@paleo/openclaw-channel-mock-core": "0.2.2",
-    "@paleo/openclaw-discord-mock": "0.2.2",
-    "@paleo/openclaw-slack-mock": "0.2.2"
+    "@paleo/openclaw-channel-mock-core": "0.2.3",
+    "@paleo/openclaw-discord-mock": "0.2.3",
+    "@paleo/openclaw-slack-mock": "0.2.3"
   },
   "devDependencies": {
     "@types/node": "~24.12.4",
diff --git a/packages/openclaw-qa-runner/templates/Dockerfile b/packages/openclaw-qa-runner/templates/Dockerfile
index 90fabb8..c6a26c5 100644
--- a/packages/openclaw-qa-runner/templates/Dockerfile
+++ b/packages/openclaw-qa-runner/templates/Dockerfile
@@ -7,22 +7,10 @@
 ARG QA_RUNNER_BASE_TAG
 FROM paleo/openclaw-qa-runner-base:${QA_RUNNER_BASE_TAG}
 
-# Stage the host-resolved @paleo/openclaw-* packages into an in-image staging dir,
-# then rewrite the consumer `package.json`'s absolute host `file:` URIs to point
-# at the staged copies before `npm install`. Host install must use `--install-links`
-# so `node_modules/@paleo/*` are materialized copies (not symlinks).
-COPY --chown=claw:claw package.json package-lock.json* /opt/qa-src/
-COPY --chown=claw:claw node_modules/@paleo/openclaw-channel-mock-core/ /opt/qa-staging/openclaw-channel-mock-core/
-COPY --chown=claw:claw node_modules/@paleo/openclaw-discord-mock/ /opt/qa-staging/openclaw-discord-mock/
-COPY --chown=claw:claw node_modules/@paleo/openclaw-slack-mock/ /opt/qa-staging/openclaw-slack-mock/
-COPY --chown=claw:claw node_modules/@paleo/openclaw-qa-runner/ /opt/qa-staging/openclaw-qa-runner/
+COPY --chown=claw:claw package.json package-lock.json /opt/qa-src/
 COPY --chown=claw:claw openclaw.json /home/claw/.openclaw/openclaw.json
 
-# The host lockfile pins integrity against the original host paths; it cannot
-# survive the URI rewrite, so drop it and let npm regenerate.
-RUN node -e "const fs=require('fs');const p='./package.json';const j=JSON.parse(fs.readFileSync(p,'utf8'));for(const k of Object.keys(j.dependencies||{})){if(k.startsWith('@paleo/openclaw-')){const n=k.slice('@paleo/'.length);j.dependencies[k]='file:/opt/qa-staging/'+n;}}fs.writeFileSync(p,JSON.stringify(j,null,2));" && \
-    rm -f package-lock.json && \
-    npm install --include=dev --install-links && \
+RUN npm ci --include=dev && \
     OPENCLAW_CONFIG_PATH=/home/claw/.openclaw/openclaw.json npx openclaw plugins registry --refresh
 
 # Consumer customizations below. Add RUN/COPY/ENV as needed.
diff --git a/packages/openclaw-slack-mock/CHANGELOG.md b/packages/openclaw-slack-mock/CHANGELOG.md
index 949cb98..00b5344 100644
--- a/packages/openclaw-slack-mock/CHANGELOG.md
+++ b/packages/openclaw-slack-mock/CHANGELOG.md
@@ -1,5 +1,13 @@
 # @paleo/openclaw-slack-mock
 
+## 0.2.3
+
+### Patch Changes
+
+- Improved documentation
+- Updated dependencies
+  - @paleo/openclaw-channel-mock-core@0.2.3
+
 ## 0.2.2
 
 ### Patch Changes
diff --git a/packages/openclaw-slack-mock/README.md b/packages/openclaw-slack-mock/README.md
index b962804..17fde18 100644
--- a/packages/openclaw-slack-mock/README.md
+++ b/packages/openclaw-slack-mock/README.md
@@ -1,10 +1,45 @@
 # @paleo/openclaw-slack-mock
 
-Synthetic Slack-shaped OpenClaw channel plugin. Registers as channel `slack-mock` with a restricted action surface (`read`, `edit`, `delete`, `react`, `reactions`, `search`). No `send` / `thread-create` / `thread-reply`. Bare-channel inbounds auto-thread: the first agent outbound creates a thread anchored on the inbound message id; every subsequent outbound from that turn lands in the same thread.
+Synthetic Slack-shaped OpenClaw channel plugin. Registers as channel `slack-mock`. Restricted Slack-shaped action surface: `read`, `edit`, `delete`, `react`, `reactions`, `search`. No `send` / `thread-create` / `thread-reply`. Bare-channel inbounds auto-thread: the first agent outbound creates a thread anchored on the inbound message id; every subsequent outbound from the same turn lands in that thread.
 
-Backed by [`@paleo/openclaw-channel-mock-core`](../openclaw-channel-mock-core/) with `surface: "slack"` and `autoThread: true`. Pair with [`@paleo/openclaw-qa-runner`](../openclaw-qa-runner/) for the QA harness.
+Backed by [`@paleo/openclaw-channel-mock-core`](https://www.npmjs.com/package/@paleo/openclaw-channel-mock-core) (`surface: "slack"`, `autoThread: true`). Pair with [`@paleo/openclaw-qa-runner`](https://www.npmjs.com/package/@paleo/openclaw-qa-runner) for the QA harness.
 
-`Provider` / `Surface` / `OriginatingChannel` on inbound metadata are claimed as `slack-mock` so the SDK routes tool-schema discovery to this plugin.
+## Install
+
+```sh
+npm i -D @paleo/openclaw-slack-mock
+```
+
+The runner depends on this package transitively — installing `@paleo/openclaw-qa-runner` already pulls it in.
+
+## Enable
+
+In your `openclaw.json`:
+
+```json
+{
+  "plugins": {
+    "load": { "paths": ["/opt/qa-src/node_modules/@paleo/openclaw-slack-mock"] },
+    "entries": { "slack-mock": { "enabled": true } }
+  },
+  "channels": {
+    "slack-mock": {
+      "baseUrl": "http://bus:43123",
+      "botUserId": "openclaw",
+      "botDisplayName": "OpenClaw QA",
+      "allowFrom": ["*"]
+    }
+  }
+}
+```
+
+`enabled: true` must be **static**. Auto-enable for `origin: "config"` plugins is timing-sensitive against the plan-resolution `explicitlyEnabled` check.
+
+## Target format
+
+Canonical destination is the `to` param. Accepts `channel:<id>` / bare `<id>` / `dm:<id>` / `group:<id>` / `thread:<channelId>/<threadId>`.
+
+`Provider` / `Surface` / `OriginatingChannel` on inbound metadata are claimed as `slack-mock` so the SDK routes tool-schema discovery to this plugin. The `chat_id` envelope shape is **not** rewritten — assert on `conversation.id` / `threadId`.
 
 ## Attribution
 
diff --git a/packages/openclaw-slack-mock/package.json b/packages/openclaw-slack-mock/package.json
index f932a38..af46bd1 100644
--- a/packages/openclaw-slack-mock/package.json
+++ b/packages/openclaw-slack-mock/package.json
@@ -1,6 +1,6 @@
 {
   "name": "@paleo/openclaw-slack-mock",
-  "version": "0.2.2",
+  "version": "0.2.3",
   "description": "Synthetic Slack-shaped OpenClaw channel plugin for automated QA scenarios.",
   "keywords": [
     "openclaw",
@@ -63,7 +63,7 @@
     "openclaw": "*"
   },
   "dependencies": {
-    "@paleo/openclaw-channel-mock-core": "0.2.2"
+    "@paleo/openclaw-channel-mock-core": "0.2.3"
   },
   "devDependencies": {
     "@types/node": "~24.12.4",