Plugins

A generative-UI surface (ADR 0038): the agent renders HTML / Markdown / charts / Mermaid diagrams / SVG / React on demand into a sandboxed iframe — the way Claude Artifacts and Open WebUI do it. React artifacts can import a curated offline set (d3, chart.js, lucide) and the protoLabs design-system @pl/ui wrappers. The shell page is served by this plugin and iframed; the agent's generated code renders in a nested sandbox with no access to the console. It also versions generated FILES (ADR 0092): save_file_artifact stores a .docx/.xlsx/.pptx/ .pdf/image as a download artifact with a diffable text preview and (for images) a thumbnail.

Execution-grounded code-solve (ADR 0064) — the verifier-grounded coder for VERIFIABLE coding tasks. Contributes a `coder_solve` tool + a `coder` subagent that turn "implement X, here are the tests" into a TEST-VERIFIED solution via a difficulty-gated search ladder (greedy → best-of-k + execution-select → tree-search refine-on-failing-tests → fusion), gated on the tests actually passing — not an LLM judge. Composes the `delegates` plugin (candidate generation over an openai/acp delegate) + a subprocess pytest runner (the verifier); it does not reimplement the spawn primitive. With no tests it degrades to a single un-verified candidate — it shines on verifiable work.

Engineering rituals as skills: /grill (relentless plan interview), /standup (operational status report), /code-review (adversarial review-workflow driver, ADR 0077), /due-diligence (adopt/build/defer verdict via the due-diligence workflow), /writing-skills (the skill-authoring discipline), and adr-authoring (the house ADR/plan-doc conventions — agent-retrievable, so agent-authored ADRs meet the bar) — plus a skill_writer subagent that drafts SKILL.md skills to that discipline. Prompt-only: no tools, routes, network, or filesystem access. The grilling, code-review, and skill-authoring material is adapted from mattpocock/skills (MIT).

A hot-swappable registry of the agents and endpoints your agent can talk to — fleet A2A agents, OpenAI-compatible endpoints, ACP coding agents — managed from the console.

Answer questions about protoAgent from its own documentation. Ships a keyword index over the bundled docs + `docs_search` / `docs_read` agent tools + a skill that teaches search → read → cite. A console Docs reader view and ⌘K search come next. First-party, on by default; disable via `plugins.disabled: [docs]`.

A Python interpreter the agent runs code in — its headline use is programmatic tool-calling (call several tools, compose the results in code, return only what matters). It runs arbitrary model-authored code: a scrubbed-env subprocess with a hard timeout is isolation, not a true sandbox (code can reach disk/network as the server user), so it ships disabled — for a trusted model or hardened container only.

The agent records its own friction — missing or awkward tools, confusing errors, wrong paths it recognizes — alongside auto-captured tool errors and escape-hatch reaches. One ledger, split into harness vs model friction, so the improvement backlog writes itself from real use. A self-report pattern in the spirit of NousResearch/hermes-agent. Bundled example — opt in via plugins.enabled.

A minimal starting point — one tool, one bundled skill, and a console view. Copy it to scaffold your own plugin by hand.

A single shared markdown note the agent and the operator both read and write. The flagship reference plugin (ADR 0034 S4) — it owns its storage, its agent tools, and a live-preview markdown editor it serves itself (ADR 0038), with CodeMirror vendored so it runs offline. Replaces the legacy native Notes surface.

Live diagram of the agent fleet — every agent a node, a directed edge wherever one agent can delegate to another. Crawls outward from this agent's delegates (using the peer tokens it already holds), enriched with each peer's public card + health.

Build your own plugins — the authoring kit + reference plugin. A scaffold tool, an architect subagent, the building-plugins skill, and a console guide, in one bundle.

Run your agent as a Telegram bot — inbound DMs and groups, replies back out. A tiny reference communication plugin.

Declarative, multi-step subagent workflows (ADR 0002) — chain subagent steps (some in parallel) into repeatable recipes. Extracted from core to an opt-in plugin (lean core); the engine taps core via the plugin SDK (graph.sdk.run_subagent), never core internals.

Board-driven coding-orchestration plugin for protoAgent: beads board + ACP spawn loop + planning layer + console view. Install via plugin install <git-url>.

Your career coach & job-hunt copilot — strategy, mock interviews, honest CV feedback, and offer/negotiation coaching, plus an autonomous evaluate→tailor→apply pipeline. A full-surface plugin showcase.

Read AND write GitHub over the gh CLI — PRs, issues, diffs, CI, repo files; plus gated create/edit/merge/close, comments, labels (per-agent write gating). Ships a read-only Issues/PRs board, a util-bar New-issue widget, and the user-only /issue command.

An autonomous SpaceTraders fleet commander — crew subagents, navigation/mining/trading tools, workflows, and a Fleet dashboard. The reference full-bundle plugin.

Portfolio Manager agent for the protoLabs.studio autonomous team

Multi-team orchestration plugin for protoAgent — one PM across many project boards over A2A

Google Workspace for the agent, built to grow. Ships Gmail (read + draft) and Calendar (read) today — gmail_list_unread,

Browser-automation plugin for protoAgent, backed by vercel-labs/agent-browser: tools + skill + workflows + a live browser panel. plugin install <git-url>.

Full Linear integration in one plugin — six linear_* tools (list teams/issues, search, get, create with dedup, comment),

Drive the Palmier Pro macOS video editor from protoAgent over a warm, persistent MCP session

MythxEngine — a protoAgent plugin: design a band, run it in a generated survival world. Mirror of protoLabsAI/mythxengine-sdk:integrations/protoagent-plugin (auto-published on tag).

A full terminal (xterm.js + a real PTY over WebSocket) as a protoAgent console plugin — themed from the protoAgent design system.

Trading research as a full bundle — market data, backtesting, factor IC, a behavioral journal, a gated paper broker, the research desk, and a Quant Desk dashboard.

But can it run DOOM? A protoAgent console plugin — play shareware DOOM in a console view via WebAssembly (prboom). No DOS layer.

Let the lead agent spawn a CLI coding agent — protoCLI, Claude Code, Codex, or Gemini CLI — and hand it a real coding job over the Agent Client Protocol.

Run your agent as a Discord bot — an inbound DM/channel gateway plus outbound posting tools.