research-pipeline-runner

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Research Pipeline Runner

研究管道运行器

Goal: let a user trigger a full pipeline with one natural-language request, while keeping the run auditable (Units + artifacts + checkpoints).

This skill is coordination:

semantic work is done by the relevant skills’
```
SKILL.md
```
scripts are deterministic helpers (scaffold/validate/compile), not the author

目标：让用户通过一个自然语言请求触发完整管道，同时保持运行过程可审计（包含Units、工件与检查点）。

本Skill的核心是协调调度：

语义层面的工作由对应Skill的
```
SKILL.md
```
完成
脚本是确定性的辅助工具（用于搭建/验证/编译），而非内容生成主体

Inputs

输入

User goal (one sentence is enough), e.g.:
- “给我写一个 agent 的 latex-survey”
Optional:
- explicit pipeline path (e.g.,
```
pipelines/arxiv-survey-latex.pipeline.md
```
  )
- constraints (time window, language: EN/中文, evidence_mode: abstract/fulltext)

用户目标（一句话即可），例如：
- “给我写一个关于agent的latex-survey”
可选参数：
- 明确的管道路径（如
```
pipelines/arxiv-survey-latex.pipeline.md
```
  ）
- 约束条件（时间范围、语言：EN/中文、证据模式：摘要/全文）

Outputs

输出

A workspace under
```
workspaces/<name>/
```
containing:
- ```
STATUS.md
```
  ,
```
GOAL.md
```
  ,
```
PIPELINE.lock.md
```
  ,
```
UNITS.csv
```
  ,
```
CHECKPOINTS.md
```
  ,
```
DECISIONS.md
```
- pipeline-specific artifacts (papers/outline/sections/output/latex)

```
workspaces/<name>/
```
目录下的工作区，包含：
- ```
STATUS.md
```
  、
```
GOAL.md
```
  、
```
PIPELINE.lock.md
```
  、
```
UNITS.csv
```
  、
```
CHECKPOINTS.md
```
  、
```
DECISIONS.md
```
- 管道专属工件（论文/大纲/章节/输出内容/latex文件）

Non-negotiables

不可违背规则

Use
```
UNITS.csv
```
as the execution contract; one unit at a time.
Respect checkpoints (
```
CHECKPOINTS.md
```
): no long prose until required approvals are recorded in
```
DECISIONS.md
```
(survey default:
```
C2
```
).
Stop at HUMAN checkpoints and wait for explicit sign-off.
Never create workspace artifacts in the repo root; always use
```
workspaces/<name>/
```
.

将
```
UNITS.csv
```
作为执行契约；每次仅执行一个单元。
严格遵循检查点（
```
CHECKPOINTS.md
```
）：未获得明确批准（记录在
DECISIONS.md
中）不得撰写长篇文稿（默认调研管道的检查点为
```
C2
```
）。
遇到HUMAN检查点时暂停，等待明确的批准指令。
禁止在仓库根目录创建工作区工件；必须使用
```
workspaces/<name>/
```
目录。

Decision tree: pick a pipeline

决策树：选择管道

User goal → choose:

Survey/综述/调研 + Markdown draft →
```
pipelines/arxiv-survey.pipeline.md
```

Survey/综述/调研 + PDF output →

pipelines/arxiv-survey-latex.pipeline.md

Snapshot/速览 →
```
pipelines/lit-snapshot.pipeline.md
```
Tutorial/教程 →
```
pipelines/tutorial.pipeline.md
```
Systematic review/系统综述 →
```
pipelines/systematic-review.pipeline.md
```
Peer review/审稿 →
```
pipelines/peer-review.pipeline.md
```

根据用户目标选择对应管道：

调研/综述/调研 + Markdown草稿 →
```
pipelines/arxiv-survey.pipeline.md
```

调研/综述/调研 + PDF输出 →

pipelines/arxiv-survey-latex.pipeline.md

速览 →
```
pipelines/lit-snapshot.pipeline.md
```
教程 →
```
pipelines/tutorial.pipeline.md
```
系统综述 →
```
pipelines/systematic-review.pipeline.md
```
审稿 →
```
pipelines/peer-review.pipeline.md
```

Recommended run loop (skills-first)

Strict-mode behavior (by design)

严格模式行为（设计如此）

--strict

runs, several semantic C3/C4 artifacts are treated as scaffolds until explicitly marked refined. This is intentional: it prevents bootstrap JSONL from silently passing into C5 writing (a major source of hollow/templated prose).

Create these markers only after you have manually refined/spot-checked the artifacts:

```
outline/subsection_briefs.refined.ok
```
```
outline/chapter_briefs.refined.ok
```
```
outline/evidence_bindings.refined.ok
```
```
outline/evidence_drafts.refined.ok
```
```
outline/anchor_sheet.refined.ok
```
```
outline/writer_context_packs.refined.ok
```

The runner may BLOCK even if the JSONL exists; add the marker after refinement, then rerun/resume the unit.

Finish:

merge → audit → (optional) LaTeX scaffold/compile

在

--strict

运行模式下，多个语义层面的C3/C4工件会被视为脚手架，直到被明确标记为已优化。这是有意设计的：防止引导式JSONL文件无声进入C5撰写阶段（这是导致内容空洞/模板化的主要原因）。

仅在手动优化/抽查工件后，创建以下标记文件：

```
outline/subsection_briefs.refined.ok
```
```
outline/chapter_briefs.refined.ok
```
```
outline/evidence_bindings.refined.ok
```
```
outline/evidence_drafts.refined.ok
```
```
outline/anchor_sheet.refined.ok
```
```
outline/writer_context_packs.refined.ok
```

即使JSONL文件存在，运行器也可能会阻塞；优化完成后添加标记文件，然后重新运行/恢复单元。

完成阶段：

合并 → 审计 → （可选）LaTeX脚手架搭建/编译

Optional CLI helpers (debug only)

可选CLI辅助工具（仅用于调试）

Kickoff + run (optional; convenient, not required):

python scripts/pipeline.py kickoff --topic "<topic>" --pipeline <pipeline-name> --run --strict

Resume:

python scripts/pipeline.py run --workspace <ws> --strict

Approve checkpoint:

python scripts/pipeline.py approve --workspace <ws> --checkpoint C2

Mark refined unit:

python scripts/pipeline.py mark --workspace <ws> --unit-id <U###> --status DONE --note "LLM refined"

启动+运行（可选；便捷操作，非必须）：

python scripts/pipeline.py kickoff --topic "<topic>" --pipeline <pipeline-name> --run --strict

恢复运行：

python scripts/pipeline.py run --workspace <ws> --strict

批准检查点：

python scripts/pipeline.py approve --workspace <ws> --checkpoint C2

标记已优化单元：

python scripts/pipeline.py mark --workspace <ws> --unit-id <U###> --status DONE --note "LLM refined"

Handling common blocks

常见阻塞问题处理

HUMAN approval required: summarize produced artifacts, ask for approval, then record it and resume.
Quality gate blocked (
```
output/QUALITY_GATE.md
```
exists): treat current outputs as scaffolding; refine per the unit’s
```
SKILL.md
```
; mark
```
DONE
```
; resume.
No network: use offline imports (
```
papers/imports/
```
or
```
arxiv-search --input
```
).
Weak coverage: broaden queries or reduce/merge subsections (
```
outline-budgeter
```
) before writing.

需要HUMAN批准：总结已生成的工件，请求批准，记录批准信息后恢复运行。
质量门阻塞（存在
```
output/QUALITY_GATE.md
```
）：将当前输出视为脚手架；根据对应单元的
```
SKILL.md
```
进行优化；标记为
```
DONE
```
后恢复运行。
无网络：使用离线导入（
```
papers/imports/
```
或
```
arxiv-search --input
```
）。
覆盖范围不足：在撰写前扩大查询范围或合并/缩减小节（使用
```
outline-budgeter
```
）。

Quality checklist

质量检查清单

```
UNITS.csv
```
statuses reflect actual outputs (no
```
DONE
```
without outputs).
No prose is written unless
```
DECISIONS.md
```
explicitly approves it.
The run stops at HUMAN checkpoints with clear next questions.
In strict mode, scaffold/stub outputs do not get marked
```
DONE
```
without refinement.

```
UNITS.csv
```
中的状态与实际输出一致（无输出不得标记为
```
DONE
```
）。
仅当
```
DECISIONS.md
```
中明确批准后，才可撰写文稿。
遇到HUMAN检查点时暂停，并明确提示下一步操作。
严格模式下，脚手架/占位输出未经过优化不得标记为
```
DONE
```
。