# research_copilot **Repository Path**: frontxiang/research_copilot ## Basic Information - **Project Name**: research_copilot - **Description**: No description available - **Primary Language**: Unknown - **License**: MIT - **Default Branch**: main - **Homepage**: None - **GVP Project**: No ## Statistics - **Stars**: 0 - **Forks**: 0 - **Created**: 2026-10-14 - **Last Updated**: 2026-10-14 ## Categories & Tags **Categories**: Uncategorized **Tags**: None ## README # Research Copilot English | [简体中文](README.zh-CN.md) Research Copilot is a Trellis-style, research-native, multi-platform CLI (`rc`) for running academic research as a controlled, stateful workflow. It models each piece of work as a task with a generic lifecycle (`planning -> in_progress -> verify -> completed`) crossed with a research activity `kind` (literature, ideation, experiment, writing, polish, review, rebuttal). Steering is **injection-driven**: each turn a coding-agent hook runs `rc context`, injecting the current workflow state plus a deterministic next-step recommendation computed from the task graph — you decide what to do next, nothing is auto-created. This rebuild **supersedes the previous Claude Code plugin architecture** (320+ skills / MCP servers shipped as a plugin). See [the redesign spec](docs/superpowers/specs/2026-06-05-research-copilot-trellis-redesign-design.md) for the locked decisions and rationale. - Author: ldm2060 - Repo: https://github.com/ldm2060/research_copilot ## Status **Phase 0 — shipped:** - `packages/core` — the pure engine: task model, lifecycle FSM, research-state recommender, verify checks, workflow parser, context builder. - `packages/cli` — the `rc` command-line tool. - `packages/adapters` — the platform registry plus the Claude Code configurator. - `research-kit/` — workflow.md, 10 neutral `rc-*` agent templates, config defaults, spec templates. - **Claude Code is the one fully shipped platform** (class-1 push injection via a `UserPromptSubmit` hook). Other platforms are designed in the adapter registry; their adapters land in later phases (see the matrix below). ## Installation ### Quick Start (npx - no installation) ```bash npx @research-copilot/cli init --user your-name --claude ``` This automatically installs both the CLI and the plugin package. ### NPM (Recommended) ```bash npm install -g @research-copilot/cli ``` ### Alternative Methods - **pnpm**: `pnpm add -g @research-copilot/cli` - **Yarn**: `yarn global add @research-copilot/cli` - **From Source**: Clone and build (see [INSTALLATION.md](./INSTALLATION.md)) - **GitHub Releases**: Download pre-built archives (see [Releases](https://github.com/ldm2060/research_copilot/releases)) See [INSTALLATION.md](./INSTALLATION.md) for detailed installation instructions, platform-specific notes, and troubleshooting. **Note:** The `rc init` command automatically installs `@research-copilot/plugin`, which provides 6 research workflow skills and 10 agents for Claude Code and other supported platforms. ## Quick Start Guide New to Research Copilot? Start here: - **[Getting Started Guide](docs/usage/getting-started.md)** — Step-by-step tutorial from installation to your first task - **[快速入门指南(中文)](docs/usage/getting-started.zh-CN.md)** — 中文版快速入门教程 ## The `rc` commands | Command | What it does | |---|---| | `rc init -u [--claude]` | Scaffold `.research/` and (with `--claude`) wire Claude Code config. | | `rc task create --kind --title [--venue ] [--parent

]` | Create a research task; prints its id and sets it active. | | `rc task start ` | Move a task `planning -> in_progress`. | | `rc task set-status ` | Apply a lifecycle transition explicitly. | | `rc task verify ` | Run the verify gate; failure rolls the task back to `in_progress`. | | `rc task complete ` | Move a task `verify -> completed`. | | `rc task add-gap --desc --suggest ` | Record an open gap that drives next-step recommendations. | | `rc task current` | Print the active task id. | | `rc context [--platform

] [--inject] [--format text\|json]` | Emit the per-turn injection block; called by the Claude Code hook each turn. | | `rc doctor` | Check `.research/`, `workflow.md`, and `.claude/settings.json` exist; exit 1 on failure. | `kind` is one of: `literature`, `ideation`, `experiment`, `writing`, `polish`, `review`, `rebuttal`. Full flag reference and examples: [docs/usage/commands.md](docs/usage/commands.md). ## Skills Research Copilot provides 6 high-level skills for orchestrating research workflows: | Skill | Description | When to Use | |---|---|---| | `/full-research-workflow` | Complete pipeline (literature → submission) | Starting new research project | | `/literature-search` | Focused paper search + baseline locking | Finding baselines for topic | | `/experiment-design` | Design and launch experiments | Running validation experiments | | `/paper-polish` | De-AI and style refinement | After writing first draft | | `/submission-sprint` | Iterative review-fix loop | Pre-submission optimization | | `/sanity-check` | 6-dimension final audit | Before final submission | See [docs/skills.md](docs/skills.md) for detailed usage. ## Platform support matrix | Platform | Status | Injection | |---|---|---| | Claude Code | ✅ Complete | Class-1 push — `UserPromptSubmit` hook -> `rc context` | | Codex | ✅ Complete | Class-1 push — `UserPromptSubmit` hook (version-gated) | | OpenCode | ✅ Complete | Class-1 push — in-process `chat.system.transform` plugin | | Gemini CLI | ✅ Complete | Class-1 push — `BeforeAgent` hook | | Cursor | ✅ Complete | Class-2 breadcrumb — sessionStart once + always-on `Active task:` rule | | Windsurf | ✅ Complete | Class-2 breadcrumb — always-on rule (agent-less) | | Kiro / Qoder / CodeBuddy / Droid / Pi / Copilot / Kilo / Antigravity | 🔄 Planned | Registry + template increment | **Class-1** platforms support per-turn push injection; **class-2** platforms cannot, so the agent re-echoes an `Active task:` breadcrumb each turn and re-resolves state. See [docs/dev/architecture.md](docs/dev/architecture.md) and [docs/dev/adding-a-platform.md](docs/dev/adding-a-platform.md). ## Relationship to the old plugin Research Copilot was previously a Claude Code plugin (a marketplace bundle of 320+ skills, 10 agents, 6 Python MCP servers, and a SessionStart guard hook). That architecture is **superseded** by this TypeScript-full-stack, multi-platform rebuild. The decision log (D1–D8) and migration path are recorded in [the redesign spec](docs/superpowers/specs/2026-06-05-research-copilot-trellis-redesign-design.md) and [docs/dev/adr/0001-trellis-emulation.md](docs/dev/adr/0001-trellis-emulation.md). ## Documentation - **Usage (researchers):** [docs/usage/](docs/usage/README.md) — command reference, Claude Code setup, a full workflow walkthrough. - **Development (contributors):** [docs/dev/architecture.md](docs/dev/architecture.md), [docs/dev/core-api.md](docs/dev/core-api.md), [docs/dev/adding-a-platform.md](docs/dev/adding-a-platform.md), [docs/dev/testing.md](docs/dev/testing.md), [docs/dev/adr/0001-trellis-emulation.md](docs/dev/adr/0001-trellis-emulation.md).