Crypto Research Agent Skills

简体中文 · Getting started · Capabilities · Architecture

Portable, evidence-gated Agent Skills for cryptography, zero-knowledge proofs, and lattice research. The repository includes one skill-creation meta-skill and fifteen research skills spanning idea triage, literature, novelty, formal review, parameters, experiments, paper writing, venue-specific writing style, and adversarial pre-review.

The suite targets the working discipline expected around leading cryptography and security venues. It does not guarantee global novelty, proof correctness, protocol security, reproducibility on untested systems, acceptance, authorship decisions, or autonomous submission. It forbids fabricated citations, proof steps, experiments, and reviewer outcomes.

Quick Start

Requirements: Git and Python 3.10 or later. The installer uses only the Python standard library.

git clone https://github.com/RiaoMike/crypto-research-agent-skills.git
cd crypto-research-agent-skills
python scripts/validate.py skills

Install all skills for your harness:

# Codex
python scripts/install.py --target codex

# Claude Code
python scripts/install.py --target claude

# Shared Agent Skills directory
python scripts/install.py --target agents

# DeepSeek or another custom harness
python scripts/install.py --target custom --dest /path/to/your/harness/skills

The installer refuses to overwrite an existing skill unless you explicitly pass --force. Use --dry-run to inspect the operation first, and repeat --skill NAME to install selected skills.

Start a Research Project

Open a fresh agent session after installation and ask:

Use crypto-research-pipeline to evaluate this ZKP/lattice idea as a gated research project.
Search and verify prior art before claiming novelty. Separate definitions, theorems,
proofs, reductions, protocol-security claims, parameters, estimates, measurements,
and reproduced results. Stop or pivot if the evidence defeats the idea. Produce a
research-state ledger, next falsifiable actions, and only evidence-supported paper text.

[Insert idea and constraints here]

The orchestrator routes work through the specialized skills and preserves blockers. It stops for accountable human decisions on ethics, disclosure, authorship, final claims, and external submission.

Skill Groups

• Meta: crypto-research-skill-creator
• Orchestration: crypto-research-pipeline
• Discovery: crypto-idea-evaluator, crypto-literature-search, crypto-paper-reader, crypto-related-work-synthesis, crypto-novelty-check
• Formal and domain review: crypto-theorem-review, zkp-security-review, lattice-parameter-audit
• Evidence and publication: crypto-experiment-reproducibility, crypto-paper-writing, crypto-paper-review
• Venue writing style: crypto-venue-paper-writing, ccs-writing-style, usenixsec-writing-style

See the capability map for the exact output contracts and boundaries.

Harness Guides

• Codex
• Claude Code
• DeepSeek-based harnesses
• Generic Agent Skills clients

The core layout follows the open Agent Skills specification. Claude Code's documented personal and project skill locations are covered in its official skills guide.

Develop and Validate

python -m unittest discover -s tests -v
python scripts/validate.py skills
python scripts/install.py --target custom --dest .install-smoke --dry-run

Use crypto-research-skill-creator when adding or refactoring a skill. Every research skill must retain explicit trigger, scope, inputs, success criteria, failure modes, workflow, output contract, integrity rules, and evaluation scenarios.

License

MIT. See LICENSE.