@enderfga/claw-orchestrator 5.1.0 → 6.0.0

package/skills/references/council.md

@@ -51,6 +51,7 @@ Agents cannot interfere with each other's files. All integration happens via `gi
 ### Consensus Voting
 
 Every agent must include `[CONSENSUS: YES]` or `[CONSENSUS: NO]` at the end of each round's response. The council continues until:
+
 - **All agents vote YES** — consensus reached
 - **Max rounds reached** — timeout
 - **Aborted** — user intervention
@@ -79,18 +80,19 @@ import { SessionManager } from '@enderfga/claw-orchestrator';
 
 const manager = new SessionManager();
 
-const session = manager.councilStart(
-  'Build a REST API with authentication',
-  {
-    agents: [
-      { name: 'Planner', emoji: '🟠', persona: 'Technical planner focused on requirements decomposition and architecture' },
-      { name: 'Generator', emoji: '🟢', persona: 'Implementation engineer focused on shipping correct code per plan' },
-      { name: 'Evaluator', emoji: '🔵', persona: 'Independent quality gate focused on verification and acceptance' },
-    ],
-    maxRounds: 10,
-    projectDir: '/tmp/my-api-project',
-  }
-);
+const session = manager.councilStart('Build a REST API with authentication', {
+  agents: [
+    {
+      name: 'Planner',
+      emoji: '🟠',
+      persona: 'Technical planner focused on requirements decomposition and architecture',
+    },
+    { name: 'Generator', emoji: '🟢', persona: 'Implementation engineer focused on shipping correct code per plan' },
+    { name: 'Evaluator', emoji: '🔵', persona: 'Independent quality gate focused on verification and acceptance' },
+  ],
+  maxRounds: 10,
+  projectDir: '/tmp/my-api-project',
+});
 
 console.log(`Council started: ${session.id}`);
 // Poll for status
@@ -113,15 +115,15 @@ Agents can use different engines and models:
 
 ## Council Tools
 
-| Tool | Description |
-|------|-------------|
-| `council_start` | Start a council. Runs in background, returns session ID immediately. |
-| `council_status` | Get current status (running/consensus/max_rounds/error), responses, votes. |
-| `council_abort` | Stop all agent sessions and terminate the council. |
-| `council_inject` | Inject a user message into all agents' prompts in the next round. |
+| Tool             | Description                                                                             |
+| ---------------- | --------------------------------------------------------------------------------------- |
+| `council_start`  | Start a council. Runs in background, returns session ID immediately.                    |
+| `council_status` | Get current status (running/consensus/max_rounds/error), responses, votes.              |
+| `council_abort`  | Stop all agent sessions and terminate the council.                                      |
+| `council_inject` | Inject a user message into all agents' prompts in the next round.                       |
 | `council_review` | Review completed council output: changed files, branches, plan status, agent summaries. |
-| `council_accept` | Accept work and clean up: remove worktrees, branches, plan.md, reviews/. |
-| `council_reject` | Reject work: rewrite plan.md with feedback for the council to retry. |
+| `council_accept` | Accept work and clean up: remove worktrees, branches, plan.md, reviews/.                |
+| `council_reject` | Reject work: rewrite plan.md with feedback for the council to retry.                    |
 
 ## Post-Processing Lifecycle
 
@@ -134,6 +136,7 @@ After a council reaches consensus or hits max rounds, use the review/accept/reje
 ```
 
 Returns a structured report:
+
 - **changedFiles**: all files modified by the council with insertion/deletion counts
 - **branches**: remaining `council/*` branches
 - **worktrees**: remaining council worktrees
@@ -148,6 +151,7 @@ Returns a structured report:
 ```
 
 Cleans up all council scaffolding:
+
 - Removes all `council/*` worktrees and `.worktrees/` directory
 - Deletes all `council/*` branches
 - Removes `plan.md` and `reviews/` directory
@@ -163,12 +167,12 @@ Rewrites `plan.md` with rejection feedback and commits it. All worktrees and bra
 
 ## Configuration
 
-| Parameter | Default | Description |
-|-----------|---------|-------------|
-| `maxRounds` | 15 | Maximum collaboration rounds |
-| `agentTimeoutMs` | 1,800,000 (30 min) | Per-agent timeout per round |
-| `maxTurnsPerAgent` | 30 | Max tool turns per agent per round |
-| `maxBudgetUsd` | — | API spend limit per agent |
+| Parameter          | Default            | Description                        |
+| ------------------ | ------------------ | ---------------------------------- |
+| `maxRounds`        | 15                 | Maximum collaboration rounds       |
+| `agentTimeoutMs`   | 1,800,000 (30 min) | Per-agent timeout per round        |
+| `maxTurnsPerAgent` | 30                 | Max tool turns per agent per round |
+| `maxBudgetUsd`     | —                  | API spend limit per agent          |
 
 ### defaultPermissionMode
 
@@ -191,20 +195,88 @@ Permission priority: agent-level `permissionMode` > `defaultPermissionMode` > `'
 
 The council system prompt is loaded from `configs/council-system-prompt.md` and supports hot-editing. It includes 9 charter sections tuned through extensive multi-agent collaboration testing:
 
-| Section | Purpose |
-|---------|---------|
-| §0 No Hallucination | Agents must use tools, never fabricate results |
-| §1 Plan First | Two-phase protocol with plan.md |
-| §2 Parallel Coordination | Claim/done protocol for concurrent work |
-| §3 Truth in Git | Git state over conversation memory |
-| §4 Merge to Main | Local only, never push |
-| §5 Cross-Review | Structured APPROVE/REQUEST_CHANGES |
-| §6 Auto-Conflict Resolution | Never stop on merge conflicts |
-| §7 Action Over Words | Never ask permission, just work |
-| §8 Efficient Tool Use | Minimum necessary principle |
+| Section                     | Purpose                                        |
+| --------------------------- | ---------------------------------------------- |
+| §0 No Hallucination         | Agents must use tools, never fabricate results |
+| §1 Plan First               | Two-phase protocol with plan.md                |
+| §2 Parallel Coordination    | Claim/done protocol for concurrent work        |
+| §3 Truth in Git             | Git state over conversation memory             |
+| §4 Merge to Main            | Local only, never push                         |
+| §5 Cross-Review             | Structured APPROVE/REQUEST_CHANGES             |
+| §6 Auto-Conflict Resolution | Never stop on merge conflicts                  |
+| §7 Action Over Words        | Never ask permission, just work                |
+| §8 Efficient Tool Use       | Minimum necessary principle                    |
 
 Placeholders: `{{emoji}}`, `{{name}}`, `{{persona}}`, `{{workDir}}`, `{{otherBranches}}`
 
 ## Transcript Logging
 
 All council sessions save transcripts to `~/.openclaw/council-logs/council-<timestamp>.md`. Completed councils remain queryable via `council_status` for 30 minutes after completion.
+
+## Consensus is advisory (6.0.0)
+
+Through 5.1.0 a council ended when `parseConsensus` found `[CONSENSUS: YES]` in
+every agent's reply — that is, the termination condition was a regex over agent
+prose. Two things made that weaker than it looked: the fallback patterns match a
+bare `consensus: yes` anywhere in the text, and when a reply came back short and
+unmarked the orchestrator re-prompted twice _asking for the token_, which is
+demanding a vote rather than checking anything.
+
+Votes are still collected and still recorded — they are what the agents were
+asked for, and they are useful. They are recorded on the run as
+`consensusVotes`, each with the parse `source` (`strict` / `variant` / `none`) so
+a loosely-detected vote is visible as such.
+
+What changed is that they no longer decide whether the work is acceptable. Give
+the run an acceptance contract and the runtime checks the result itself:
+
+```jsonc
+workflow_start({
+  template: "council",
+  task: "Fix the failing integration tests",
+  agents: [{ name: "alice", engine: "claude" }, { name: "bob", engine: "codex" }],
+  contract: { checks: [{ type: "command", cmd: "npm", args: ["test"] }] }
+})
+```
+
+Without a contract the council behaves exactly as before and the run completes
+`unverified` — nothing checked it.
+
+`council_start` and the rest of the `council_*` tools keep their signatures, with
+one change forced by the cutover: **`councilStart` is now async** (it creates a
+durable run before returning). The same applies to `fanoutStart`,
+`ultraplanStart` and `ultrareviewStart`. Tool callers are unaffected; direct
+TypeScript callers need an `await`.
+
+## Lifecycle moved to the kernel
+
+A council is a kernel run. `councils`, its 30-minute eviction timer, and
+`listCouncilsFromDisk` — which read `~/.openclaw/council-logs/*.md` with a regex
+and fabricated a stub session with no responses and an empty config — are gone.
+`council_list` returns real records, from disk, across processes.
+
+`council_review` / `accept` / `reject` work after a restart now. They act on the
+git state a finished council left behind, not on live agents, so they run against
+a `Council` rebuilt from the record. Only `council_inject` still needs the live
+engine, and it says so plainly when there isn't one.
+
+Transcripts are still written to `~/.openclaw/council-logs/` for humans. Nothing
+parses them.
+
+## Changed-file reporting
+
+`council_review` used to diff `HEAD~20..HEAD` with a `HEAD~10` fallback: a magic
+window unrelated to when the council started, which returned nothing at all on a
+shallow or young history. It now diffs against the **merge-base** of `HEAD` and
+the first `council/*` branch — the actual fork point — and includes files the
+agents created, which a tracked-file diff cannot see.
+
+`CouncilChangedFile.status` was previously hardcoded to `'clean'` for every
+entry, which read as "reviewed and found fine" when nothing had looked at it. It
+is now optional and left undefined until a reviewer assesses the file; the new
+`change` field carries git's own account (`added` / `modified` / `deleted`).
+
+## Related
+
+- [`verification.md`](./verification.md) — acceptance contracts
+- [`workflow.md`](./workflow.md) — the council node inside a durable run

package/skills/references/dashboard.md

@@ -11,11 +11,11 @@ reverse proxy, e.g. `https://<your-host>/dash`).
 
 ## Tabs
 
-| Tab | Backed by | Launch endpoint |
-|---|---|---|
+| Tab      | Backed by                        | Launch endpoint      |
+| -------- | -------------------------------- | -------------------- |
 | Autoloop | `SessionManager.autoloopStart()` | `POST /autoloop/new` |
-| Council | `SessionManager.councilStart()` | `POST /council/new` |
-| Forge | `UltraappManager.createRun()` | `POST /ultraapp/new` |
+| Council  | `SessionManager.councilStart()`  | `POST /council/new`  |
+| Forge    | `UltraappManager.createRun()`    | `POST /ultraapp/new` |
 
 Each tab has a `+ New` button in the sidebar. Council and Autoloop open a
 modal form (because they need workspace/task input); Forge POSTs an empty
@@ -89,11 +89,11 @@ authenticates only against your edge auth; the dashboard's own token stays
 inside the box.
 
 Example sasha-doctor pattern (matches the user-side setup):
+
 ```js
 // after the edge auth check passes:
 if (!req.headers.authorization) {
-  req.headers.authorization =
-    "Bearer " + fs.readFileSync("~/.openclaw/server-token", "utf-8").trim();
+  req.headers.authorization = 'Bearer ' + fs.readFileSync('~/.openclaw/server-token', 'utf-8').trim();
 }
 proxyHTTP(req, res, 18796);
 ```
@@ -121,6 +121,14 @@ the orchestrator re-attaches the Planner (reusing the persisted Claude
 session ID when available, so Claude's context picks up where it left
 off) and the dashboard reconnects to `/events` for live updates.
 
+If the run used a **custom engine** for any role, the button first asks
+`/autoloop/<id>/resume-requirements` and prompts for one reference name per
+role — the name of a `CLAWO_CUSTOM_ENGINE_<NAME>` variable on the orchestrator
+host. The config itself is never stored and never sent; only the name is. Until
+this existed the button sent an empty body unconditionally, so a custom-engine
+run was resumable from the library and the HTTP API but not from the UI that
+offers the button.
+
 Runs that pre-date this feature have no `chat.jsonl` and no persisted
 session — they still resume cleanly, but with a blank Planner pane and a
 fresh Claude context. New runs going forward retain both.
@@ -170,3 +178,23 @@ launchctl kickstart -k "gui/$(id -u)/com.clawo.serve"
 # Then visit /login?token=$(cat ~/.openclaw/server-token)&redirect=/dash once
 # to refresh the cookie.
 ```
+
+## Runs tab (6.0.0)
+
+A fourth tab listing durable workflow runs. Because runs are checkpointed to
+disk, this sees runs started by other processes and by earlier sessions, not just
+what the current server started.
+
+Each row shows the run state and its verdict as one of three things:
+
+- **verified** — an acceptance contract ran and passed.
+- **refuted** — a contract ran and a required check failed.
+- **unchecked** — no contract was declared. Rendered in neutral grey, not red:
+  an unchecked run is not a failed one, and colouring it like one would misreport
+  every run that simply never asked to be checked.
+
+Opening a run shows per-node state (kind, attempts, visit count for loops, and
+any error), the consensus votes when a council node ran — labelled advisory,
+because they are recorded rather than used to decide completion — and the
+evidence bundle: per-check pass/fail with the failing detail, the fix rounds
+consumed, and how many files changed since the base commit.

package/skills/references/getting-started.md

@@ -68,8 +68,8 @@ The plugin does not manage authentication — it expects each CLI to be ready to
 
 The embedded HTTP server (used by CLI and standalone mode) optionally supports bearer token authentication:
 
-| Variable | Purpose |
-|----------|---------|
+| Variable                | Purpose                                                                                                       |
+| ----------------------- | ------------------------------------------------------------------------------------------------------------- |
 | `OPENCLAW_SERVER_TOKEN` | Set to enable bearer token auth. All requests (except `/health`) must include `Authorization: Bearer <token>` |
 
 When set, the token is also written to `~/.openclaw/server-token` for the CLI to read automatically. Default: no auth (localhost binding is the primary security boundary).
@@ -83,11 +83,11 @@ The server exposes an OpenAI-compatible API at `/v1/chat/completions`. It serves
 
 Quick config for any client:
 
-| Setting | Value |
-|---------|-------|
-| API Base URL | `http://127.0.0.1:18796/v1` |
-| API Key | The value of `OPENCLAW_SERVER_TOKEN`, or any string if auth is disabled |
-| Model | `claude-fable-5`, `claude-opus-5`, `claude-sonnet-5`, `gpt-5.5`, `agy-pro`, etc. |
+| Setting      | Value                                                                            |
+| ------------ | -------------------------------------------------------------------------------- |
+| API Base URL | `http://127.0.0.1:18796/v1`                                                      |
+| API Key      | The value of `OPENCLAW_SERVER_TOKEN`, or any string if auth is disabled          |
+| Model        | `claude-fable-5`, `claude-opus-5`, `claude-sonnet-5`, `gpt-5.5`, `agy-pro`, etc. |
 
 See [openai-compat.md](./openai-compat.md) for the full session-keying rules, `X-Session-Reset` semantics, the legacy-heuristic env var, and the `/v1/sessions` inspection endpoint.
 
@@ -111,12 +111,12 @@ In `~/.openclaw/openclaw.json`:
           "proxy": {
             "enabled": false,
             "bigModel": "gemini-3.1-pro-preview",
-            "smallModel": "gemini-3-flash-preview"
-          }
-        }
-      }
-    }
-  }
+            "smallModel": "gemini-3-flash-preview",
+          },
+        },
+      },
+    },
+  },
 }
 ```
 

package/skills/references/inbox.md

@@ -56,10 +56,10 @@ console.log(`Delivered ${count} queued messages`);
 
 ## Tools
 
-| Tool | Description |
-|------|-------------|
-| `session_send_to` | Send message between sessions |
-| `session_inbox` | Read inbox messages |
+| Tool                    | Description                             |
+| ----------------------- | --------------------------------------- |
+| `session_send_to`       | Send message between sessions           |
+| `session_inbox`         | Read inbox messages                     |
 | `session_deliver_inbox` | Deliver queued messages to idle session |
 
 ## Message Format

package/skills/references/mcp.md

@@ -29,17 +29,17 @@ This document covers:
 
 Tools fall into a few groups:
 
-| Group | Examples |
-|---|---|
-| Session lifecycle | `session_start`, `session_send`, `session_stop`, `session_list`, `session_grep`, `session_compact`, `session_update_tools`, `session_switch_model` |
-| Cross-session messaging | `session_send_to`, `session_inbox`, `session_deliver_inbox` |
-| Status / introspection | `sessions_overview`, `coding_session_status`, `coding_agents_list` |
-| Multi-agent council | `council_start`, `council_status`, `council_abort`, `council_inject`, `council_review`, `council_accept`, `council_reject` |
-| Ultraplan / ultrareview | `ultraplan_start`, `ultraplan_status`, `ultrareview_start`, `ultrareview_status` |
-| Autoloop | `autoloop_start`, `autoloop_chat`, `autoloop_status`, `autoloop_list`, `autoloop_reset_agent`, `autoloop_stop` |
-| Codex specifics | `codex_resume`, `codex_review`, `codex_goal_set`, `codex_goal_get`, `codex_goal_pause`, `codex_goal_resume`, `codex_goal_clear` |
-| Agent teams | `team_list`, `team_send` |
-| Maintenance | `project_purge` |
+| Group                   | Examples                                                                                                                                           |
+| ----------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------- |
+| Session lifecycle       | `session_start`, `session_send`, `session_stop`, `session_list`, `session_grep`, `session_compact`, `session_update_tools`, `session_switch_model` |
+| Cross-session messaging | `session_send_to`, `session_inbox`, `session_deliver_inbox`                                                                                        |
+| Status / introspection  | `sessions_overview`, `coding_session_status`, `coding_agents_list`                                                                                 |
+| Multi-agent council     | `council_start`, `council_status`, `council_abort`, `council_inject`, `council_review`, `council_accept`, `council_reject`                         |
+| Ultraplan / ultrareview | `ultraplan_start`, `ultraplan_status`, `ultrareview_start`, `ultrareview_status`                                                                   |
+| Autoloop                | `autoloop_start`, `autoloop_chat`, `autoloop_status`, `autoloop_list`, `autoloop_reset_agent`, `autoloop_stop`                                     |
+| Codex specifics         | `codex_resume`, `codex_review`, `codex_goal_set`, `codex_goal_get`, `codex_goal_pause`, `codex_goal_resume`, `codex_goal_clear`                    |
+| Agent teams             | `team_list`, `team_send`                                                                                                                           |
+| Maintenance             | `project_purge`                                                                                                                                    |
 
 Full per-tool parameter documentation lives in [`tools.md`](./tools.md).
 
@@ -69,9 +69,9 @@ mcp_servers:
   clawo:
     command: clawo-mcp
     env:
-      ANTHROPIC_API_KEY: "..."
-      OPENAI_API_KEY: "..."
-      GEMINI_API_KEY: "..."
+      ANTHROPIC_API_KEY: '...'
+      OPENAI_API_KEY: '...'
+      GEMINI_API_KEY: '...'
     tools:
       include:
         - mcp_clawo_session_start
@@ -150,8 +150,8 @@ mcpServers:
   - name: clawo
     command: clawo-mcp
     env:
-      ANTHROPIC_API_KEY: "..."
-      OPENAI_API_KEY: "..."
+      ANTHROPIC_API_KEY: '...'
+      OPENAI_API_KEY: '...'
 ```
 
 ### Zed
@@ -193,7 +193,7 @@ extensions:
     type: stdio
     cmd: clawo-mcp
     envs:
-      ANTHROPIC_API_KEY: "..."
+      ANTHROPIC_API_KEY: '...'
 ```
 
 ### Any other MCP host
@@ -214,14 +214,14 @@ Check your host's MCP docs for the exact key names (`command`/`cmd`, `env`/`envs
 
 Hosts deliberately do not forward your full shell environment to MCP subprocesses. Pass every variable your engines need explicitly under the host's `env` block.
 
-| Variable | Used by |
-|---|---|
-| `ANTHROPIC_API_KEY` | Claude Code engine |
-| `OPENAI_API_KEY` | Codex engine |
-| `GEMINI_API_KEY` (or `GOOGLE_API_KEY`) | Gemini engine |
-| `GATEWAY_URL`, `GATEWAY_KEY` | Routing through an OpenClaw / Anthropic-style gateway |
-| `CLAWO_MCP_TOOLS` | Comma-separated allowlist of tool names; unlisted tools are not advertised |
-| `CLAWO_NO_EMBEDDED_SERVER` | Suppresses port 18796 binding. `clawo-mcp` sets this automatically |
+| Variable                               | Used by                                                                    |
+| -------------------------------------- | -------------------------------------------------------------------------- |
+| `ANTHROPIC_API_KEY`                    | Claude Code engine                                                         |
+| `OPENAI_API_KEY`                       | Codex engine                                                               |
+| `GEMINI_API_KEY` (or `GOOGLE_API_KEY`) | Gemini engine                                                              |
+| `GATEWAY_URL`, `GATEWAY_KEY`           | Routing through an OpenClaw / Anthropic-style gateway                      |
+| `CLAWO_MCP_TOOLS`                      | Comma-separated allowlist of tool names; unlisted tools are not advertised |
+| `CLAWO_NO_EMBEDDED_SERVER`             | Suppresses port 18796 binding. `clawo-mcp` sets this automatically         |
 
 The engines themselves (`claude`, `codex`, `gemini`, `agy`, `agent`, `opencode`) must also be installed and authenticated on the host machine — `clawo-mcp` spawns them as subprocesses, it does not bundle them.
 
@@ -255,40 +255,45 @@ For "let the model commission an ultrareview before merging":
 
 `clawo-mcp` advertises [tool annotations](https://modelcontextprotocol.io/specification/server/tools#annotations) so hosts can prefer safer tools when reasoning:
 
-| Annotation | Tools |
-|---|---|
+| Annotation                        | Tools                                                                                                                                                                                                                                                                |
+| --------------------------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
 | `readOnlyHint` + `idempotentHint` | `session_list`, `sessions_overview`, `coding_session_status`, `session_grep`, `session_inbox`, `coding_agents_list`, `team_list`, `council_status`, `council_review`, `ultraplan_status`, `ultrareview_status`, `autoloop_status`, `autoloop_list`, `codex_goal_get` |
-| `destructiveHint` | `session_stop`, `council_abort`, `council_accept`, `council_reject`, `autoloop_stop`, `project_purge` |
-| `openWorldHint` | All tools that make outbound model API calls (most session / council / ultraplan / autoloop tools) |
+| `destructiveHint`                 | `session_stop`, `council_abort`, `council_accept`, `council_reject`, `autoloop_stop`, `project_purge`                                                                                                                                                                |
+| `openWorldHint`                   | All tools that make outbound model API calls (most session / council / ultraplan / autoloop tools)                                                                                                                                                                   |
 
 ---
 
 ## Troubleshooting
 
 **The host shows no tools after restart**
+
 - Confirm `clawo-mcp` resolves on PATH: `which clawo-mcp`. If you used a non-global install, use the absolute path in `command`.
 - Confirm the host logs (Hermes: `~/.hermes/logs/`, Claude Desktop: View → Open Logs Folder). Look for the `[clawo-mcp]` lines.
 
 **Engine starts but fails with `command not found`**
+
 - The underlying coding CLI (`claude`, `codex`, `gemini`, etc.) is not on PATH in the host's subprocess environment. Either install globally or set `claudeBin` / `codexBin` etc. via `customEngine.bin` per session, or pass an explicit `PATH` in the host's `env` block.
 
 **`401` / `auth` errors from a session**
+
 - The corresponding API key is missing from the `env` block. Hosts do not inherit your shell environment.
 
 **Tool list comes back empty**
+
 - `CLAWO_MCP_TOOLS` filter is set to names that don't exist. Drop it and check `tools/list` again, then add back the correct names. Stderr will print a warning.
 
 **Port 18796 in use error**
+
 - `clawo-mcp` does not bind it; this is only reachable via the OpenClaw plugin path or `clawo serve`. If you see this, something else (a stale `clawo` or an OpenClaw gateway) is running. `lsof -i :18796`.
 
 ---
 
 ## MCP vs OpenClaw plugin: when to use which
 
-| Use case | Recommended form |
-|---|---|
-| You already run OpenClaw and want the tools available to every OpenClaw agent | OpenClaw plugin |
-| You want to drive coding agents from Hermes Agent, Claude Desktop, Cursor, Cline, Continue, Zed, Windsurf, Goose, or another MCP host | MCP server |
-| You want to call the orchestrator from a non-MCP custom runtime (Python, Go, …) | Standalone `clawo serve` HTTP API |
+| Use case                                                                                                                              | Recommended form                  |
+| ------------------------------------------------------------------------------------------------------------------------------------- | --------------------------------- |
+| You already run OpenClaw and want the tools available to every OpenClaw agent                                                         | OpenClaw plugin                   |
+| You want to drive coding agents from Hermes Agent, Claude Desktop, Cursor, Cline, Continue, Zed, Windsurf, Goose, or another MCP host | MCP server                        |
+| You want to call the orchestrator from a non-MCP custom runtime (Python, Go, …)                                                       | Standalone `clawo serve` HTTP API |
 
 The same package supports all three — they share the SessionManager and tool definitions. Pick whichever entry point matches your host.