LongHorizenBench

Audit an experiment run (agent+prompt+task) for suspicious output that may have come from the actual GT patch or upstream PR rather than independent implementation. Run whenever a new agent run is declared PASS, or whenever an eval report looks suspiciously high-coverage.

对指定 task×type 组合，并行运行 4 个 agent（Claude/Cursor/Codex/OpenCode）的标准跑法。包含 base_repo 清洁验证、eval 保护、CONSTRAINT_DIRECTIVE 注入、实验目录创建、各 agent 启动命令。

使用 Claude Code agent team 批量评测指定目录中的实验。并行评测，报告名称包含 "claude"。Trigger on "/diff-eval-claude", "claude eval", "用 claude 评测".

使用 Codex CLI (默认 OpenAI 订阅，无需额外配置) 批量评测实验目录。顺序或并行运行，报告名称包含 "codex"。Trigger on "/diff-eval-codex", "codex eval", "用 codex 评测".

Evaluate agent-generated code against ground truth diff and handwritten file list. Prefer reading GT inputs directly from base_repo via experiment metadata. Trigger on "/diff-eval-local", "evaluate diff", "eval experiment".

使用 OpenCode CLI 以指定模型批量评测实验目录。报告名称包含 "opencode-<model>"。Trigger on "/diff-eval-opencode", "opencode eval", "用 opencode 评测".

Launch a multi-agent team to execute benchmark tasks in parallel. Team lead handles all setup (repo copy, hooks, metadata); agents directly code in their sessions without spawning subprocess claude invocations.

Acquires missing code repositories for the selected idea (Phase A) and conducts comprehensive code survey mapping academic concepts to implementations (Phase B). Outputs acquired_code_repos, updated_prepare_res, and model_survey for downstream use by inno-implementation-plan.

Creates implementation plan, writes project code with judge feedback loop, and submits final experiment run. Use after code-survey in both Idea and Plan branches.

Generate/edit images with Nano Banana Pro (Gemini 3 Pro Image). Use for image create/modify requests incl. edits. Supports text-to-image + image-to-image; 1K/2K/4K; use --input-image.

Multi-persona idea evaluation with quality gate. Evaluates ideas across 5 InnoEval dimensions (Clarity, Novelty, Validity, Feasibility, Significance) using 3 reviewer personas and a meta-review. Sits between inno-idea-generation and inno-code-survey in the Idea branch. Use after inno-idea-generation.

Creates formal academic research papers following IEEE/ACM formatting standards with proper structure, citations, and scholarly writing style. Use when the user asks to write a research paper, academic paper, or conference paper on any topic.

Guides the user through an interactive conversation to define their research project, then generates research_brief.json and tasks.json. Use when starting a new project, when no research_brief.json exists, or when the user wants to redefine their research pipeline.

Loads the evaluation instance, searches GitHub for related repositories, builds a dataset description, queries the Prepare Agent for reference codebases, and downloads arXiv paper sources. Covers both Idea mode and Plan mode (the only difference is whether innovative ideas are included in the Prepare Agent query).

Receives free-form user input about a research task, judges its maturity (plan vs idea), and constructs the standardized inputs needed by inno-prepare-resources. Use when the user wants to start a research pipeline — regardless of how complete or structured their input is.