Boardwalk plugins

Official plugins that let agent harnesses drive Boardwalk: shared skills for the boardwalk CLI, plus the hosted control-plane MCP server.

Five shared skills for Claude Code, Codex, and Cursor: boardwalk-overview (the platform mental model), boardwalk-use-cli (the first-party CLI: scaffold, run, validate, deploy, trigger, operate), write-good-workflows (authoring quality), write-good-loops (iterating agent loops), and equip-agents (skills, tools, MCP, memory, and human-input gates inside a workflow). OpenClaw and OpenCode currently wire the boardwalk-use-cli skill. For Claude Code the plugin also connects the remote MCP server so the model can create, schedule, trigger, and monitor workflows directly.

Layout

The canonical plugin payload (skills + Codex manifest) lives under plugins/boardwalk/ so the Codex marketplace installer (npx codex-plugin add) finds it where it expects. Per-harness manifests for the other three harnesses live at the repo root and reference the same shared skills:

• .claude-plugin/ — Claude Code plugin + marketplace manifest
• .mcp.json — remote MCP server config (Claude Code auto-loads it when the plugin is enabled)
• .cursor-plugin/ — Cursor plugin manifest
• .agents/plugins/ — Codex marketplace catalog
• openclaw.plugin.json — OpenClaw plugin manifest
• plugins/boardwalk/.codex-plugin/ — Codex plugin manifest
• plugins/boardwalk/skills/ — shared skill definitions (all harnesses point here)

OpenCode needs no manifest: it loads Agent Skills natively, so you link the boardwalk-use-cli skill in directly (see OpenCode under Install).

The skills are single-source-of-truth: a change lands in every harness at once, with no copy step.

Install

Claude Code

claude plugin marketplace add boardwalk-labs/plugins
claude plugin install boardwalk@boardwalk-labs

Codex

npx codex-plugin add boardwalk-labs/plugins

Then open /plugins in Codex and enable boardwalk.

Cursor

Cursor Marketplace publication is pending. Until then, install from a local checkout by symlinking the repo into your Cursor plugins directory:

ln -s "$(pwd)" ~/.cursor/plugins/local/boardwalk

OpenClaw

openclaw plugins install ./

OpenCode

OpenCode has its own plugin system, but it is a different shape: OpenCode plugins are JavaScript/TypeScript hook modules (custom tools and lifecycle hooks) listed under plugin in opencode.json, not skill bundles, so the Boardwalk skill does not ship as an OpenCode plugin. Instead OpenCode loads Agent Skills natively and reads the same skill with nothing to install. From a checkout of this repo, link the skill into OpenCode's skills directory:

mkdir -p ~/.config/opencode/skills
ln -s "$(pwd)/plugins/boardwalk/skills/boardwalk-use-cli" \
  ~/.config/opencode/skills/boardwalk-use-cli

If you already installed the Claude Code plugin, OpenCode also discovers skills under ~/.claude/skills/, so the Boardwalk skill is available there with no extra step.

Remote MCP server

For Claude Code, the plugin also bundles the Boardwalk control-plane MCP server (.mcp.json, loaded automatically when the plugin is enabled). It talks to https://api.boardwalk.sh/mcp/v1 and gives the model the full control plane: create and update workflows, add cron / rate / one-shot schedules, trigger runs, tail run output, answer human-input gates, and manage environments, variables, and secrets metadata.

Auth is a Boardwalk API key sent as a bearer header. Create one in the web console (Settings > API keys), then:

export BOARDWALK_API_KEY=bwk_...

Restart Claude Code (or run /mcp and reconnect) after setting it. Without the variable the skills still work; only the MCP connection needs it.

Any MCP-capable harness can connect the same server manually: it is a plain HTTP MCP server at https://api.boardwalk.sh/mcp/v1, authenticated with an Authorization: Bearer $BOARDWALK_API_KEY header. Point your harness's MCP config at that URL and header; from the Claude Code CLI (outside the plugin), for example:

claude mcp add --transport http boardwalk https://api.boardwalk.sh/mcp/v1 \
  --header "Authorization: Bearer $BOARDWALK_API_KEY"

What it does

Installs five skills. boardwalk-overview orients a model that is new to Boardwalk: what the platform is and how workflows, triggers, and runs fit together. boardwalk-use-cli gives the model the boardwalk CLI surface: scaffolding (init), validating (check), building the artifact (build), authenticating, deploying and triggering (deploy / run / cancel), inspecting runs and usage (runs / usage), and managing workflows, secrets, environments, variables, and inference providers, plus project linking, auth precedence, the run-event channels, and self-hosting knobs. write-good-workflows covers authoring quality: the run function and its typed I/O, the workflow.jsonc descriptor, SDK primitives, run legibility, efficiency, and surviving restarts. write-good-loops teaches the model to author an agent loop that iterates until a goal is reached: the core loop shape, the layered exits every loop needs (verifier, iteration cap, budget, no-progress), bounded-vs-recurring topology, maker/checker verification, durable cross-run state, and not paying to wait. equip-agents covers giving an agent() call skills, inline tools, MCP servers, memory, and human-input gates. The CLI itself ships separately as @boardwalk-labs/cli.

The Boardwalk repos

• boardwalk — the open-source single-node engine: cron scheduling, webhooks, durable runs, run history
• sdk-typescript — @boardwalk-labs/workflow, the TypeScript API a workflow program imports
• cli — boardwalk: scaffold, validate, deploy, run
• examples — copyable workflow templates (boardwalk init --template)
• runner — self-hosted runner: your machines execute hosted-scheduled runs

Hosted platform and docs: boardwalk.sh.

License

MIT — see LICENSE.