AutoResearchLab

Validate task output against criteria before calling Finish. Use when the task has a validation spec (criteria, optionalChecks). Load this skill, write output to sandbox, run validate script, fix issues if validation fails, then call Finish.

Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, update or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.

End-to-end git release workflow for this repository. Use when the user asks to publish a release, says release or ship, or otherwise wants Codex to complete release work by checking and updating docs, reviewing all changes since the previous tag, committing and pushing changes, choosing a version tag based on release scope, and drafting concise Chinese GitHub-ready release notes in a fenced Markdown block.

Develop complete, runnable software projects. Use this whenever the user asks to write code, create an app, or analyze data with scripts. This skill enforces persisting all code to disk (via the WriteFile tool which runs inside the Docker workspace) and executing it (via the RunCommand tool), rather than outputting scattered Markdown code snippets.

Advanced RAG-style research schema with literature-grounded fields. Use when RefineIdea for hardware-aware or citation-heavy proposals.

Quality criteria for refined_idea output. Use before RefineIdea or when ValidateRefinedIdea returns low score. Focus on concrete, decomposable, and specific output.

Three academic research templates (Heilmeier, scientific discovery, system optimization) with schemas and use cases. Use when RefineIdea to align output structure with research type.

Guidelines for refining broad or fuzzy ideas into concrete research topics. Use when user input is vague to guide ExtractKeywords and RefineIdea. Includes RESEARCH_TOPIC_SCHEMA and refinement tactics.

Guidelines for extracting arXiv search keywords from fuzzy research ideas. Use before ExtractKeywords or when EvaluatePapers suggests retry. Ensures keywords are technical, domain-specific, and suitable for literature retrieval.

Guidelines for citing literature in refined ideas. Use [Source ID: X] format when referencing papers. Apply when RefineIdea output should be grounded in retrieved papers.

Criteria for evaluating paper relevance (EvaluatePapers) and filtering (FilterPapers). Use before EvaluatePapers/FilterPapers to ensure consistent scoring and selection. Score 1–5, should_retry when score < 3.

Common patterns for decomposing research tasks. Use when decomposing complex tasks to ensure MECE (mutually exclusive, collectively exhaustive) splits and clear phase boundaries. Essential for Decompose tool. Apply when CheckAtomicity returns atomic=false.

Guidelines for scoping research ideas into actionable tasks. Use when the root idea is broad or ambiguous to ensure appropriate granularity and feasibility. Helps with initial decomposition. Apply at root (task "0") and when refining broad sub-tasks.

Guidelines for defining input/output specifications of atomic tasks. Use when formatting tasks (FormatTask) to ensure Task Agent and validator have clear, checkable criteria. Apply when CheckAtomicity returns atomic=true.

Synthesize literature, findings, or multiple sources into a coherent report. Use when task involves literature review, synthesizing research, or combining multiple inputs into themes and conclusions.

Ensure research reports, comparison reports, and synthesis documents include proper citations and a References section. Use when task output is a Markdown report with quantitative claims, benchmarks, or findings that require source attribution for credibility. Load alongside markdown-reporter or web-research for citation-heavy tasks.

Criteria for judging if a task is atomic (executable in one step). Use with CheckAtomicity to decide when to Decompose vs FormatTask. Essential for atomicity decisions. Apply when processing each task in the planning workflow.

Rules for task dependencies in decomposition. Use when defining dependencies in Decompose/AddTasks to ensure acyclic, sibling-only, correct execution order. Critical for valid plan structure.

Analyze structured data (JSON, tables, metrics) and produce insights. Use when task involves analyzing data, computing metrics, or deriving conclusions from structured input. Output can be report (Markdown) or structured (JSON). Covers aggregation, comparison, trends, and validation.