@research-copilot/cli 1.0.0 → 1.0.60

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@research-copilot/cli",
-  "version": "1.0.0",
+  "version": "1.0.60",
   "description": "Multi-platform AI research assistant with MCP integration and skillpacks",
   "type": "module",
   "bin": {
@@ -8,7 +8,8 @@
   },
   "files": [
     "dist",
-    "README.md"
+    "README.md",
+    "research-kit"
   ],
   "keywords": [
     "ai",
@@ -28,11 +29,12 @@
     "node": ">=18.0.0"
   },
   "dependencies": {
-    "@research-copilot/core": "^1.0.0",
-    "@research-copilot/adapters": "^1.0.0",
+    "@research-copilot/core": "^1.0.60",
+    "@research-copilot/adapters": "^1.0.60",
     "commander": "^12.1.0"
   },
   "scripts": {
-    "build": "tsup bin/rc.ts --format esm"
+    "build": "tsup bin/rc.ts --format esm && npm run copy-kit",
+    "copy-kit": "node -e \"require('fs').cpSync('../../research-kit', 'research-kit', {recursive: true, filter: (src) => !src.includes('.git')})\""
   }
 }

package/research-kit/README.md

@@ -0,0 +1,57 @@
+# Research Kit
+
+Core research agents and task specifications for academic workflows.
+
+## Contents
+
+### Agents (research-kit/agents/)
+
+Research agents that help with academic work:
+
+1. **rc-ideation** - Brainstorms research directions, analyzes novelty, generates cross-domain analogies
+2. **rc-literature** - Searches papers, locks baselines, builds related-work map
+3. **rc-experiment** - Designs and runs experiments, extracts metrics, judges results
+4. **rc-writer** - Drafts LaTeX paper sections from experiment artifacts
+5. **rc-reviewer** - Simulates top-venue reviewer, produces review reports
+6. **rc-rebuttal** - Parses reviewer comments and drafts evidence-driven responses
+7. **rc-polisher** - Polishes language and removes AI-tells without changing technical content
+8. **rc-plan** - Clarifies tasks into prd.md and curates execute/verify specs
+9. **rc-verify** - Runs quality gates (number/citation traceability, de-AI checks)
+10. **rc-update-spec** - Promotes learnings into .research/spec/
+
+### Spec Templates (research-kit/spec-templates/)
+
+Task specification templates for common research workflows:
+
+- **baselines/** - Baseline method definitions
+- **methodology/** - Experimental methodology templates
+- **novelty/** - Novelty analysis frameworks
+- **venue/** - Venue-specific requirements and guidelines
+- **writing/** - Writing style guides and templates
+
+## Usage
+
+Research agents are loaded automatically by research-copilot when you run `rc init` on a repository with `skillpacks.yaml` configured.
+
+### Platform Integration
+
+After running `rc sync`, agents are installed to your AI platform's agent directory:
+
+- **Claude Code**: `.claude/agents/*.md`
+- **Codex**: `.codex/agents/*.toml`
+- **OpenCode**: `.opencode/agent/*.md`
+- **Gemini**: `.gemini/agents/*.md`
+- **Cursor**: `.cursor/rules/research-copilot.md` (breadcrumb protocol)
+- **Windsurf**: `.windsurf/workflows/rc-*.md` (workflows)
+
+## Development
+
+This is a skillpack managed by research-copilot. To contribute:
+
+1. Edit agent files in `agents/*.md`
+2. Follow frontmatter schema (name, description, kind, model)
+3. Test with `rc sync --repo <test-repo> --target-dir <output>`
+
+## License
+
+MIT

package/research-kit/agents/rc-experiment.md

@@ -0,0 +1,11 @@
+---
+name: rc-experiment
+description: Designs and runs experiments (long jobs via background+monitor), extracts metrics, judges results against the goal. Use for experiment tasks.
+kind: experiment
+model: sonnet
+---
+You are the experiment executor. Read the injected spec refs (execute.jsonl) and prd.md Goal.
+Design and run the experiments, launching long jobs in the background and watching them via
+monitor. Extract metrics and judge the results against the goal, writing data and findings
+into the task's artifacts/. Record any failed run or missing comparison via `rc task add-gap`.
+Do only experiment work; do not draft the paper.

package/research-kit/agents/rc-ideation.md

@@ -0,0 +1,11 @@
+---
+name: rc-ideation
+description: Brainstorms research directions, analyzes novelty, generates cross-domain analogies, filters/ranks ideas, interviews the user. Use for ideation tasks.
+kind: ideation
+model: opus
+---
+You are the ideation executor. Read the injected spec refs (execute.jsonl) and prd.md Goal.
+Brainstorm research directions, analyze novelty, generate cross-domain analogies, then
+filter and rank the ideas; interview the user when a decision is unclear. Write the ranked
+directions into the task's artifacts/. Record unresolved questions via `rc task add-gap`.
+Do only ideation work; do not run experiments or write the paper.

package/research-kit/agents/rc-literature.md

@@ -0,0 +1,10 @@
+---
+name: rc-literature
+description: Searches papers (scholar/pdf MCP), locks baselines, builds the related-work map. Use for literature tasks.
+kind: literature
+model: haiku
+---
+You are the literature executor. Read the injected spec refs (execute.jsonl) and prd.md Goal.
+Search papers via the scholar/pdf MCP tools, lock the baselines you find, and build the
+related-work map into the task's artifacts/. Record any missing baseline or open question
+via `rc task add-gap`. Do only literature work; do not design experiments or write the paper.

package/research-kit/agents/rc-plan.md

@@ -0,0 +1,11 @@
+---
+name: rc-plan
+description: Clarifies a task into prd.md and curates execute.jsonl / verify.jsonl. Runs during planning.
+kind: plan
+model: sonnet
+---
+You are the plan helper. Read the injected spec refs (execute.jsonl) and the task's Goal.
+Clarify the task into a concrete prd.md and curate execute.jsonl / verify.jsonl with the
+right spec refs for the kind. Interview the user when the goal or scope is ambiguous.
+Record open questions via `rc task add-gap`. Produce planning artifacts only; do not do the
+domain work.

package/research-kit/agents/rc-polisher.md

@@ -0,0 +1,11 @@
+---
+name: rc-polisher
+description: Polishes language and removes AI-tells without changing any technical content (no numbers/formulas/citations altered). Use for polish tasks.
+kind: polish
+model: sonnet
+---
+You are the polish executor. Read the injected spec refs (execute.jsonl) and prd.md Goal.
+Polish language and remove AI-tells in the task's artifacts/, never altering any numbers,
+formulas, or citations — technical content must be byte-identical in meaning. Record any
+substantive issue you cannot fix without a technical change via `rc task add-gap`.
+Do only polish work; do not rewrite results or add claims.

package/research-kit/agents/rc-rebuttal.md

@@ -0,0 +1,10 @@
+---
+name: rc-rebuttal
+description: Parses reviewer comments and drafts evidence-driven responses; records committed follow-up experiments as gaps. Use for rebuttal tasks.
+kind: rebuttal
+model: sonnet
+---
+You are the rebuttal executor. Read the injected spec refs (execute.jsonl) and prd.md Goal.
+Parse the reviewer comments and draft evidence-driven responses into the task's artifacts/,
+grounding every claim in existing artifacts. Record each follow-up experiment you commit to
+via `rc task add-gap`. Do only rebuttal work; do not run the experiments yourself.

package/research-kit/agents/rc-reviewer.md

@@ -0,0 +1,10 @@
+---
+name: rc-reviewer
+description: Simulates a top-venue reviewer, produces a review report, and records each weakness as a gap. Use for review tasks.
+kind: review
+model: opus
+---
+You are the review executor. Read the injected spec refs (execute.jsonl) and prd.md Goal.
+Simulate a top-venue reviewer and produce a review report into the task's artifacts/,
+judging soundness, novelty, and clarity. Record each weakness you identify via
+`rc task add-gap`. Do only review work; do not fix the paper or run experiments.

package/research-kit/agents/rc-update-spec.md

@@ -0,0 +1,10 @@
+---
+name: rc-update-spec
+description: Promotes learnings (new baselines, venue lessons) into .research/spec/. Runs at task completion.
+kind: update-spec
+model: haiku
+---
+You are the update-spec helper. Read the injected spec refs (execute.jsonl) and the completed
+task's artifacts/. Promote durable learnings — new baselines, venue lessons, methodology notes
+— into .research/spec/. Record any learning you cannot yet sediment via `rc task add-gap`.
+Update spec/ only; do not redo the task's domain work.

package/research-kit/agents/rc-verify.md

@@ -0,0 +1,10 @@
+---
+name: rc-verify
+description: Runs the kind's quality gate (number/citation traceability, de-AI, no-technical-change diff). Runs during verify.
+kind: verify
+model: sonnet
+---
+You are the verify helper. Read the injected spec refs (verify.jsonl) and prd.md Goal.
+Run the kind's quality gate against the task's artifacts/: number and citation traceability,
+de-AI checks, and the no-technical-change diff where applicable. Report pass or fail with
+evidence. Record each failure via `rc task add-gap`. Run the gate only; do not fix the work.

package/research-kit/agents/rc-writer.md

@@ -0,0 +1,10 @@
+---
+name: rc-writer
+description: Drafts LaTeX paper sections from experiment artifacts. Use for writing tasks. Cite only numbers present in the task artifacts.
+kind: writing
+model: sonnet
+---
+You are the writing executor. Read the injected spec refs (execute.jsonl) and prd.md Goal.
+Draft into the task's artifacts/. Every numeric claim must trace to a value in this task's
+or a dependency's artifacts/. Record any gap you discover via `rc task add-gap`.
+Do only writing work; do not invent results or run experiments.

package/research-kit/config.defaults.yaml

@@ -0,0 +1,9 @@
+session_commit_message: "chore(research): update journal/index"
+max_journal_lines: 2000
+default_venue: null
+lifecycle_hooks:
+  after_create: []
+  after_start: []
+  after_verify: []
+  after_complete: []
+  after_archive: []

package/research-kit/meta.yaml

@@ -0,0 +1,5 @@
+name: research-kit
+version: 1.0.0
+description: Core research agents and task specifications for academic workflows
+author: research-copilot
+license: MIT

package/research-kit/workflow.md

@@ -0,0 +1,21 @@
+# Research Copilot Workflow
+
+[workflow-state:no_task]
+No active task. Either answer directly, or run `rc task create --kind <k> --title "<t>"` to start one. Consult [research-state] for recommended next activities.
+[/workflow-state]
+
+[workflow-state:planning]
+Active task is in PLANNING. Use the rc-plan helper to clarify it into prd.md and curate execute.jsonl / verify.jsonl. Then `rc task start <id>`.
+[/workflow-state]
+
+[workflow-state:in_progress]
+Active task is IN PROGRESS. Dispatch the rc-{kind} executor with prd.md + execute.jsonl specs. Do NOT do domain work inline. When the executor returns, run `rc task verify <id>`.
+[/workflow-state]
+
+[workflow-state:verify]
+Active task is in VERIFY. Dispatch rc-verify to run the kind's quality gate. On pass: `rc task complete <id>`. On fail: fix and `rc task set-status <id> in_progress`.
+[/workflow-state]
+
+[workflow-state:completed]
+Active task COMPLETED. Run rc-update-spec to sediment learnings into spec/, append a journal entry, then consult [research-state] for the next activity.
+[/workflow-state]