feat: start selected thread sessions automatically

.changeset/alcode-inline-wake-command.md

@@ -0,0 +1,5 @@
+---
+"@paleo/alcode": patch
+---
+
+Prevented OpenClaw completion wake-ups from being lost by keeping the wake command's semicolon on the `alcode` command line.

.changeset/alignfirst-merge-discipline.md

@@ -0,0 +1,5 @@
+---
+"alignfirst": patch
+---
+
+Tightened merge conflict resolution and validation guidance.

.changeset/discord-thread-rename-fidelity.md

@@ -0,0 +1,5 @@
+---
+"@paleo/openclaw-channel-mock-core": patch
+---
+
+Matched Discord thread behavior: inbound metadata uses native channel targets, `send` with `threadName` renames the resolved thread, and `thread-reply` leaves its name unchanged.

.changeset/durable-thread-handoff.md

@@ -0,0 +1,5 @@
+---
+"@paleo/alignfirst-developer-openclaw-plugin": minor
+---
+
+Added durable activation and claim handling for confirmed Slack and Discord thread starters.

.changeset/fix-pnpm-fixture-install.md

@@ -0,0 +1,5 @@
+---
+"@paleo/openclaw-test": patch
+---
+
+Fixed the pnpm fixture-install example for current pnpm releases.

.changeset/native-thread-mock-routing.md

@@ -0,0 +1,7 @@
+---
+"@paleo/openclaw-channel-mock-core": minor
+"@paleo/openclaw-slack-mock": minor
+"@paleo/openclaw-discord-mock": minor
+---
+
+Added configurable Slack thread routing, native starter receipt shapes, and canonical thread-session delivery. A send whose target names a stored thread now lands in that thread under its parent conversation, with or without an accompanying `threadId`. Discord-shaped `thread-create` now returns the native `{ ok, thread }` shape, with `partial: true` when the thread exists but its starter was not delivered; the former `threadId`, `target` and `message` fields are gone. Discord-shaped `thread-reply` now accepts a bare `threadId` as its delivery target, as bundled Discord does. The test-bus fault injector accepts `threadOnly: true` to fail only a threaded send.

.changeset/scenario-context-bus-url.md

@@ -0,0 +1,5 @@
+---
+"@paleo/openclaw-test": minor
+---
+
+Added `ctx.busUrl` to `ScenarioContext` for direct test-bus calls.

.github/workflows/release.yml

@@ -91,9 +91,9 @@ jobs:
           cd "$(mktemp -d)"
           npm init -y > /dev/null
           # --omit=peer: auto-installed peers drag unrelated trees into the audit
-          for i in 1 2 3 4 5; do
+          for i in $(seq 1 10); do
             if npm install --omit=peer $pkgs; then break; fi
-            if [ "$i" = 5 ]; then echo "Install failed after 5 attempts"; exit 1; fi
+            if [ "$i" = 10 ]; then echo "Install failed after 10 attempts"; exit 1; fi
             echo "Registry not ready, retrying in 30s"
             sleep 30
           done

README.md

@@ -20,12 +20,14 @@ Nine Agent Skill stubs expose the CLI through commands such as `/alspec` in Clau
 
 ## OpenClaw Test toolkit
 
-`@paleo/openclaw-test` and three companion channel packages: they are a Dockerised regression-test harness that drives OpenClaw through synthetic Discord and Slack channels. See [packages/openclaw-test/README.md](packages/openclaw-test/README.md).
+`@paleo/openclaw-test` and three companion channel packages are a Dockerised regression-test harness that drives OpenClaw through synthetic Discord and Slack channels. See [packages/openclaw-test/README.md](packages/openclaw-test/README.md).
 
 ## AlignFirst Developer
 
 AlignFirst Developer is an AI teammate for software work, currently packaged on OpenClaw. See [alignfirst-developer.md](alignfirst-developer.md).
 
+[`@paleo/alignfirst-developer-openclaw-plugin`](packages/alignfirst-developer-openclaw-plugin/README.md) supplies its OpenClaw capabilities under plugin ID `alignfirst-developer`. Its first capability, thread handoff, durably activates the ordinary thread session after confirmed native starter delivery.
+
 ---
 
 ## Setup with your agent

alignfirst-developer-tests/.env.local.example

@@ -36,8 +36,8 @@ ALIGNFIRST_CODE_AGENT=codex
 # ALIGNFIRST_CODE_MODELS=terra
 
 # Optional: run N matrix cells concurrently, each on its own worker stack
-# (`run --parallel` wins over this).
-# OPENCLAW_TEST_PARALLEL=6
+# (`run --parallel` wins over this). A development laptop handles 3.
+OPENCLAW_TEST_PARALLEL=3
 
 # Optional overrides (defaults relative to the project dir):
 # OPENCLAW_CONFIG_PATH=./openclaw.json

alignfirst-developer-tests/Dockerfile

@@ -44,11 +44,13 @@ RUN chmod +x /opt/openclaw-test/scripts/*.mjs
 USER claw
 
 # Pin pnpm's content-addressable store outside /home/claw/projects/, which is a
-# docker volume on its own filesystem. Without this, pnpm walks up from a
-# worktree install target and creates /home/claw/projects/.pnpm-store, polluting
-# the project listing the agent grounds project names in. Cost: pnpm falls back
-# to copy instead of hardlinks across filesystems — irrelevant at fixture size.
-RUN printf 'store-dir=/home/claw/.pnpm-store\n' > /home/claw/.pnpmrc && \
+# docker volume on its own filesystem. Without this, pnpm creates a fresh
+# /home/claw/projects/.pnpm-store per cell and re-downloads every fixture
+# dependency (10 s alone, minutes under three parallel cells); with it, the
+# template install below warms the store the fixtures reuse offline. pnpm 12
+# reads this from ~/.config/pnpm/config.yaml, which `pnpm config set` writes.
+# Cost: copy instead of hardlinks across filesystems — irrelevant at fixture size.
+RUN pnpm config set store-dir /home/claw/.pnpm-store --global && \
     mkdir -p /home/claw/.pnpm-store
 
 # The @paleo/openclaw-* packages install from locally-built tarballs under
@@ -67,13 +69,13 @@ COPY --chown=claw:claw openclaw.json /home/claw/.openclaw/openclaw.json
 RUN npm ci --include=dev && \
     OPENCLAW_CONFIG_PATH=/home/claw/.openclaw/openclaw.json \
       npx openclaw plugins install --force --accept-capabilities \
-        npm:@openclaw/codex@2026.8.2 && \
+        npm:@openclaw/codex@2026.9.2 && \
     OPENCLAW_CONFIG_PATH=/home/claw/.openclaw/openclaw.json \
       npx openclaw plugins install --force --accept-capabilities \
-        npm:@openclaw/zai-provider@2026.8.2 && \
+        npm:@openclaw/zai-provider@2026.9.2 && \
     OPENCLAW_CONFIG_PATH=/home/claw/.openclaw/openclaw.json \
       npx openclaw plugins install --force --accept-capabilities \
-        npm:@openclaw/perplexity-plugin@2026.8.2 && \
+        npm:@openclaw/perplexity-plugin@2026.9.2 && \
     OPENCLAW_CONFIG_PATH=/home/claw/.openclaw/openclaw.json \
       npx openclaw plugins registry --refresh && \
     npm cache clean --force
@@ -83,7 +85,7 @@ RUN npm ci --include=dev && \
 # lumen, and external-team project orion with per-name patches.
 COPY --chown=claw:claw projects-fixture/template/ /opt/alignfirst-developer-tests/fixtures/template/
 RUN cd /opt/alignfirst-developer-tests/fixtures/template && \
-    pnpm install --frozen-lockfile --prod=false
+    pnpm install --frozen-lockfile
 
 # Install AlignFirst skills under /home/claw/.agents/skills/ (the path OpenClaw scans).
 RUN npx -y skills add https://github.com/paleo/alignfirst --global --yes \

alignfirst-developer-tests/README.md

@@ -57,9 +57,14 @@ The root and its nested `external-projects` and `lifecycle-projects` directories
 
 ## Scenarios
 
-Drop `scenarios/<id>.ts`, default-export `async (ctx: ScenarioContext) => void`. Shared helpers under `scenarios/_lib/` (skipped by the runner's discovery). Current scenarios: `A01`–`A21` and `A23`–`A26`.
+Drop `scenarios/<id>.ts`, default-export `async (ctx: ScenarioContext) => void`. Shared helpers under `scenarios/_lib/` (skipped by the runner's discovery). Current scenarios: `A01`–`A21` and `A23`–`A28`.
 
-Almost every one starts with `bootstrapThreadFromChannel` (`_lib/thread-bootstrap.ts`): it sends the channel message, waits for the starter, and asserts the channel session stopped right there — one thread post, no second one, no worktree on disk, no coding-agent call, nothing substantive leaked to the channel root. `sendInThread` then wakes the thread session, which owns the actual work. A scenario that seeds a worktree first passes its absolute path as `seededWorktreePaths` so the check still catches anything the channel session created.
+Almost every one starts with `bootstrapThreadFromChannel` (`_lib/thread-bootstrap.ts`). It sends the
+channel message, waits for exactly one confirmed native starter and one `thread_handoff start`, and
+checks the parent session's attributed tool trace for target work. The plugin wakes the thread
+session automatically; complete requests need no mechanical follow-up. `sendInThread` remains for
+genuine missing values, explicit holds, confirmations, and later requests. Target work may begin
+before the parent emits its final `NO_REPLY`, so assertions follow the starter's original cursor.
 
 `A10` exercises the real `alcode` foreground run driven as an OpenClaw background exec and rejects direct Claude or Codex launches. `A11` covers an explicit user hold. `A12` chains two delegations in one thread, exposing the heartbeat-cooldown wake gate. `A13` drives alcode directly for deterministic selected-agent new/resume coverage and Codex failure handling. The shared mock serves a bundled Codex model catalog and both agents' JSONL protocols.
 
@@ -69,6 +74,14 @@ Almost every one starts with `bootstrapThreadFromChannel` (`_lib/thread-bootstra
 
 `A23` resolves a PR URL through review and its reported outcome. `A24` carries a multi-project base refresh through one no-protocol delegation per project. `A25` captures a detailed request before workspace setup and coding. `A26` reserves the next side ticket `side-N` before workspace setup for explicit no-ticket work.
 
+`A27-human-reply-racing-startup` sends a genuine missing-ticket answer immediately after native
+starter delivery. `A28-recoverable-handoff-failure` injects one test-bus delivery failure, then
+requires one successful starter and automatic work without creating a replacement target.
+
+`A01`, `A04`, `A05` and `A21` open a thread whose starter asks for a value, then pin the
+silent seed turn (`_lib/silent-seed-turn.ts`): the thread session claims the handoff, posts nothing
+for 90 s and reads no thread history.
+
 Rebuild the CLIs and harness image before focused coverage:
 
 ```sh
@@ -80,17 +93,39 @@ ALIGNFIRST_CODE_AGENT=codex npm run e2e -- --channel all A06-off-projects A14-so
 ALIGNFIRST_CODE_AGENT=codex npm run e2e -- --channel all A17-project-creation A18-project-removal A19-project-removal-failure
 ALIGNFIRST_CODE_AGENT=codex npm run e2e -- --channel all A23-resource-url-handoff A24-multi-project-handoff A25-detailed-request-handoff A26-explicit-no-ticket
 ALIGNFIRST_CODE_AGENT=codex npm run e2e -- --channel all A10-coding-session A12-sequential-coding-sessions
+ALIGNFIRST_CODE_AGENT=codex npm run e2e -- --channel all A27-human-reply-racing-startup A28-recoverable-handoff-failure
 ALIGNFIRST_CODE_AGENT=claude npm run e2e -- --channel all A13-alcode-agent-contract A10-coding-session
 npm run e2e -- --model gpt-5.6-terra --channel all --all
 ```
 
 **Ticket-id convention:** scenario `A<S>` uses `ABC-0<S>N` (`A1` → `ABC-010`, `A2` → `ABC-020`, …; `A10` → `ABC-0100`). The mechanical mapping is a leak signal: while running `A<S>`, any `ABC-0<X>N` with `X ≠ S` is bleed from another scenario. The test sender is `ROBIN01`, listed in [`workspace/USER.md`](workspace/USER.md). A5's `aurora` is deliberately **not** a fixture name (unknown-project path).
 
-## Vendored `@paleo/openclaw-*` packages
+## Vendored packages
+
+This harness vendors the **local** sources of the four generic `@paleo/openclaw-*` packages and
+`@paleo/alignfirst-developer-openclaw-plugin`.
+The dependencies are `file:vendor/<pkg>.tgz`; [`scripts/vendor-packages.mjs`](scripts/vendor-packages.mjs)
+builds each package and `npm pack`s it into `vendor/` (gitignored). The Docker build context is this
+directory, so the tarballs must live here.
+
+`npm run env:build` chains `vendor` → `npm install` (refreshing `package-lock.json`) →
+`openclaw-test env build`, so a source edit in any of the five packages is picked up on the next
+build. Run `npm run vendor` before the first standalone `npm install`; the tarballs must exist for
+resolution.
 
-This harness always tests the **local** `@paleo/openclaw-*` sources, never npmjs — the four packages iterate in lockstep with the mocks and are frequently ahead of a publish. The dependencies are `file:vendor/<pkg>.tgz`; [`scripts/vendor-packages.mjs`](scripts/vendor-packages.mjs) (`npm run vendor`) builds each package and `npm pack`s it into `vendor/` (gitignored). The Docker build context is this dir, so the tarballs must live here — `../packages/*` is out of reach at build time.
+The plugin is explicitly allowlisted, loaded from its installed package path, and exposes optional
+tool `thread_handoff`. Slack uses `replyToMode: "off"`; Discord remains non-automatic. Both surface
+IDs map to their native receipt contract in `plugins.entries.alignfirst-developer.config.channelSurfaces`.
+
+The complementary deterministic suite makes no model calls and runs outside Docker against the
+pinned OpenClaw 2026.9.2 executable:
+
+```sh
+KEEP_THREAD_HANDOFF_ARTIFACTS=1 npm run test:integration --workspace @paleo/alignfirst-developer-openclaw-plugin --prefix ..
+```
 
-`npm run env:build` chains `vendor` → `npm install` (refreshes `package-lock.json` against the new tarballs) → `openclaw-test env build`, so a source edit in any of the four packages is picked up on the next `env:build` with no manual step. `npm pack` is byte-reproducible, so unchanged sources produce no lockfile churn. Run `npm run vendor` by hand before the first `npm install` (the tarballs must exist for it to resolve).
+It retains test-owned gateway logs, scripted-provider requests, configuration, and SQLite restart
+state under `/tmp/thread-handoff-*`. Omit the environment variable to clean fixtures automatically.
 
 ## Layout
 

alignfirst-developer-tests/openclaw.json

@@ -4,16 +4,27 @@
     "auth": { "mode": "none" }
   },
   "plugins": {
+    "allow": ["codex", "browser", "discord-mock", "slack-mock", "alignfirst-developer"],
     "load": {
       "paths": [
         "/opt/openclaw-test/src/node_modules/@paleo/openclaw-discord-mock",
-        "/opt/openclaw-test/src/node_modules/@paleo/openclaw-slack-mock"
+        "/opt/openclaw-test/src/node_modules/@paleo/openclaw-slack-mock",
+        "/opt/openclaw-test/src/node_modules/@paleo/alignfirst-developer-openclaw-plugin"
       ]
     },
     "entries": {
       "codex": { "enabled": true },
       "discord-mock": { "enabled": true },
-      "slack-mock": { "enabled": true }
+      "slack-mock": { "enabled": true },
+      "alignfirst-developer": {
+        "enabled": true,
+        "config": {
+          "channelSurfaces": {
+            "slack-mock": "slack",
+            "discord-mock": "discord"
+          }
+        }
+      }
     },
     "slots": { "memory": "none" }
   },
@@ -26,7 +37,7 @@
   "update": { "checkOnStart": false },
   "tools": {
     "profile": "coding",
-    "alsoAllow": ["message", "browser"],
+    "alsoAllow": ["message", "browser", "thread_handoff"],
     "deny": ["ask_user"]
   },
   "models": {
@@ -120,6 +131,7 @@
       "botUserId": "openclaw",
       "botDisplayName": "myclaw",
       "allowFrom": ["*"],
+      "replyToMode": "off",
       "blockStreaming": true
     }
   }