AI45Lab

AI45Lab/iDeer

Instructions file CodexOpenCode

AGENTS.md instructions for AI45Lab/iDeer, covering agents, local python environment, setup commands, installed requirements and verification.

AI45Lab/iDeer

Skill Claude CodeCodex

Bridge daily recommender ideas to research pipelines. Read history/ideas/{date}/ideas.json OR generate ideas yourself from today's scored items, then route to /idea-creator, /idea-discovery, or /research-pipeline. Use when user says '/idea-from-daily', '从今日推荐启动研究', 'pick idea from daily'.

AI45Lab/iDeer

Skill Claude CodeCodex

Use iDeer as a daily paper-reading workflow for chatbot-first users such as Codex, Gemini, or ChatGPT. Keep the original iDeer paper-digest setup, source selection, history validation, email/report/ideas workflow, but replace in-repo LLM API summarization and scoring with the current chatbot session. 适用于不用单独配置…

AI45Lab/iDeer

Skill Claude CodeCodex

Daily paper/repo digest where YOU are the reader. Fetch items from arXiv/HuggingFace/GitHub/Semantic Scholar, then read, score, summarize, and generate ideas yourself — no external LLM API calls. Use when user says '今日论文', 'daily paper', 'daily digest', '每日推荐', or wants a personalized research briefing.

AI45Lab/SAfactory

Skill Claude CodeCodex

Use this skill when helping users onboard a benchmark or environment into SAfactory, run SAfactory Docker-mode evaluation, or start SAfactory GRPO/RL training. It guides Codex to inspect the SAfactory repository, read the relevant docs, follow the established Geo3K baseline and myenv placeholder workflows, configure…

AI45Lab/Code

Agent

You are a code review agent focused on finding issues, regressions, and weak designs.

AI45Lab/Code

Agent

You are a fast exploration agent focused on understanding codebases.

AI45Lab/Code

Agent

You are a planning agent focused on designing implementation approaches.

AI45Lab/Code

Agent

You are an adversarial verification specialist. Your job is to find evidence, not to reassure.

AI45Lab/Box

Skill Claude CodeCodex

Drive the a3s-box microVM sandbox CLI - a Docker-like runtime that runs OCI/container images in hardware-isolated microVMs (libkrun). Use when the user wants to run, build, exec into, snapshot, or tear down containers with a3s-box, sandbox untrusted or agent-generated code, spin up throwaway isolated environments, or…

AI45Lab/SentrySkills

Instructions file

Instructions for AI45Lab/SentrySkills, a project described as: LLM-native skill package that teaches agents to protect themselves.

AI45Lab/SentrySkills

Skill Claude CodeCodex

SentrySkills is a workspace-local self-guard framework for AI agents.

AI45Lab/SentrySkills

Skill Claude CodeCodex

Extension layer for SentrySkills. It separates online extra-rule detection from post-model-stage knowledge management.

AI45Lab/SentrySkills

Skill Claude CodeCodex

Execute privacy and sensitive leakage guarding before output. As long as context contains sensitive information, must trigger this skill even if user only requests explanation/summary. Default to downgrade expression for tool conclusions not multi-source verified.

AI45Lab/SentrySkills

Skill Claude CodeCodex

Establish security boundaries before execution. Trigger this skill whenever involving command execution, file writing, sensitive data reading, external tool result adoption, or potential unauthorized requests; first output risk assessment and allowed/forbidden action lists.

AI45Lab/SentrySkills

Skill Claude CodeCodex

Monitor high-risk actions and behavior drift during execution. Trigger this skill when task enters command execution, tool calls, file writing, or batch modification; continuously produce event logs, alerts, and disposition recommendations.

AI45Lab/SentrySkills

Skill Claude CodeCodex

Run SentrySkills before every task using a rule-first frontend and a risk-gated model backend. The skill/framework decides sync vs async after rule gating.