菜单
Route work tasks to the appropriate sub-skill based on task context
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Reference for configuring, launching, and monitoring research iteration loops and analysis polish loops
Verifies the analysis pipeline runs end-to-end, outputs match expectations, and LaTeX compiles
Hands-on QA of the research analysis pipeline. Use whenever the user asks to verify the pipeline end-to-end, check output file integrity, validate table and figure generation, test data processing, or verify LaTeX compilation. Triggers on phrases like "test the pipeline", "QA the analysis", "check the outputs", "verify the tables", "run end-to-end", "does the paper compile", or any request to validate a research pipeline.
Define research question, hypotheses, methodology outline, and literature gap
Decompose a research question into cook-it-ready epic beads with methodology specifications
Comprehensive playbook for building rigorous, publication-quality research papers. Covers the full research output sequence from paper structure and statistical reporting through table formatting, figure design, LaTeX conventions, academic writing quality, and pre-submission review.
基于 SOC 职业分类
| name | Work |
| description | Route work tasks to the appropriate sub-skill based on task context |
| phase | work |
Generic work framework that reads task context and delegates to the right sub-skill(s). The router picks tasks, detects the work mode, and hands off to work-code, work-writing, or work-analysis.
bd ready to find available tasks (or use the task ID from arguments)bd update <id> --claimbd show <epic> for the spec, acceptance criteria table, and verification contractdrl search for relevant context from prior workRead the task title and description. Match keywords to determine the sub-skill:
| Keywords | Sub-Skill |
|---|---|
| implement, test, code, function, module, fix bug, refactor, pipeline | work-code/SKILL.md |
| write, draft, section, introduction, discussion, prose, abstract, conclusion | work-writing/SKILL.md |
| regression, analysis, table, figure, descriptive stats, robustness, estimate | work-analysis/SKILL.md |
If the task is ambiguous (no clear keyword match), use AskUserQuestion to clarify the work mode.
.claude/skills/drl/<sub-skill>/SKILL.md) and follow its instructionswork-analysis first to produce tables and figureswork-writing to draft the section referencing those outputsEvery methodological choice MUST be logged to docs/decisions/ using the ADR template (docs/decisions/0000-template.md). Use /drl:decision for guided logging. This applies regardless of which sub-skill executes the work.
bd update <id> --notes="Completed: ..."bd close <id>Gate 3: All Work Tasks Closed
| Criterion | Verification |
|---|---|
| All work tasks closed | bd list --status=in_progress is empty |
| Sub-skill gates pass | Each invoked sub-skill's verification gate passes |
| All decisions logged | ls docs/decisions/ (one ADR per decision) |
| Spec was approved | bd show <epic> confirms spec phase complete |
drl search before each task for relevant patternsdrl learn after discoveries or corrections