🚀 OpenReview Agent

Agent skill and CLI toolkit for safe OpenReview submission workflows.

• Submission Copilot • Cross-Venue Transfer • Dry-Run-First OpenReview Automation •

中文文档 • Features • Quick Start • Workflows • Safety Model • Agent Guide • Security

Important

📚 From manual forms to submission-native agents.
I was transferring two papers on OpenReview with a dozen-plus coauthors. Every venue field opened another checklist: author profile IDs, schema keys, reviewer nominations, LLM usage declarations, dataset/code links, PDFs, checklists, and conference-specific policies. After too many repeated clicks, I wondered: why is this still not agent-native?

🧭 Ten minutes with an agent.
I talked through the pain with a coding agent, and it became this project: a local tool that helps researchers inspect, plan, validate, explain, and dry-run OpenReview edits before anything is written.

🔓 Why open source it?
Many OpenReview automations live in private scripts. OpenReview Agent provides an inspectable, reproducible, dry-run-first skill and CLI layer for safer submission workflows.

🧭 Quick Navigation

Tip

I'm a human -> Continue reading this README for setup, workflows, safety boundaries, and project context.

I'm an agent -> Read SKILL.md for operating rules, status codes, command contracts, and recovery patterns.

openreview-agent is an agent skill and CLI toolkit that helps AI agents and researchers inspect OpenReview submissions, match author profiles with affiliation evidence, edit metadata safely, plan cross-venue transfers, and create dry-run-first batch submissions.

• For authors: avoid deadline-night mistakes in author IDs, venue fields, attachments, and reviewer nomination.
• For agents: use structured commands instead of ad-hoc one-off Python snippets.
• For OpenReview safety: inspect first, generate payloads second, write only with explicit confirmation.

Status: 0.1.0-alpha

Not affiliated with OpenReview, NeurIPS, ICLR, ICML, CVPR, ECCV, or any conference organization. Use only with accounts, venues, and data you are authorized to access.

⚡ Quick Start

Tell your coding agent:

Install OpenReview Agent from https://github.com/OpenClaudex/openreview-agent and set it up for safe, dry-run-first OpenReview submission workflows.

OpenReview credentials can be provided through a token, environment variables, a private local ~/.openreview.env, or the interactive prompt. Low-level command details live in SKILL.md, not in this README.

Local credential file

For repeated local use, copy the structure from docs/openreview.env.example to ~/.openreview.env, fill values locally, and restrict file permissions:

cp docs/openreview.env.example ~/.openreview.env
chmod 600 ~/.openreview.env

The CLI loads ~/.openreview.env automatically. Commit only the example structure, never real account values.

✨ Features

OpenReview Agent focuses on author-side OpenReview workflows:

• Inspects existing forum notes, signatures, readers/writers, license, content keys, and editable invitation fields.
• Matches OpenReview profiles using author name plus affiliation/history evidence.
• Fetches source submission metadata for cross-venue transfer planning.
• Maps source fields to target venue invitation schemas with enum, length, author, anonymity, and duplicate checks.
• Generates dry-run post_note_edit payloads before any write.
• Applies transfer edits only with explicit confirmation flags.
• Batch-creates independent submissions from JSON / JSONL with dry-run-first safeguards.
• Avoids default custom readers, writers, and nonreaders; venue processes should own permissions unless explicitly overridden.

🧩 Workflows

Tier	Workflow	Expected Behavior
Stable	Inspect existing submissions	Read note metadata, authors, authorids, content keys, license, and editable schema
Stable	Profile matching	Rank candidate OpenReview profiles by name and affiliation evidence
Stable	Cross-venue transfer preflight	Fetch source note, plan field mapping, validate target schema, emit dry-run payload
Stable	Safe apply	Write only after explicit confirmation, then re-check OpenReview state
Limited	Batch submissions	JSON/JSONL input, attachment handling, per-record errors, user-owned policy responsibility
Best-effort	Venue-specific custom forms	Schema fallback and manual verification recommended

🛡️ Safety Model

OpenReview Agent treats every write as high-risk.

• Dry-run by default. Transfer and batch commands do not write unless explicit apply flags are present.
• Schema-aware writes. Target invitation schema decides which content fields are allowed.
• No silent author guessing. Low-confidence author profile matches should be confirmed by a human.
• No default complex permissions. The tool does not invent readers, writers, or nonreaders by default.
• No credential persistence in the repo. Tokens/passwords must not be written to logs, payload files, screenshots, or examples; keep real values only in local private files such as ~/.openreview.env.
• No review generation or spam. The project is not a review bot and not a bulk-submission spam tool.

Read SECURITY.md before using this on private venues or live submissions.

🧪 Why This Exists

OpenReview is powerful, but author-side workflows are fragile. A small metadata mistake can mis-assign authorship, break reviewer matching, leak anonymity, miss a required declaration, or fail at the last minute because a venue-specific field changed.

OpenReview Agent is not a full reviewing system. It is a local execution layer for safer submission workflows: inspect, plan, dry-run, apply, and verify.

📚 Docs

• Agent Guide
• Security Policy
• Release Checklist
• Credential template: docs/openreview.env.example
• Venue templates: config/venues
• CLI tools: scripts/or_transfer.py, scripts/or_batch.py

🗺️ Roadmap

• v0.1: inspect, profile-match, one-command transfer dry-run/apply, batch submission dry-run/apply.
• v0.2: review-aware transfer preflight from source reviews, meta-review, and decision.
• v0.3: public review-pattern analysis by target venue and area.
• v0.4: resubmission summary-of-changes drafting with human approval.

🌐 Related Projects

OpenReview Agent focuses on author-side submission workflows. Related projects:

• OpenReview - open platform for peer review.
• openreview-py - official Python client for OpenReview.
• openreview/openreview-mcp - MCP server for openreview-py knowledge and introspection.
• OpenCodice-Research/openreview-mcp - read-oriented MCP for submissions, reviews, rebuttals, and decisions.

⭐ Star History

📄 License

MIT

---

If this project helps you avoid an OpenReview deadline mistake, please give it a ⭐ Star!

Report Issues · Feature Requests