Gitlink
/
gitlink-cli
0
0
Fork 49
Code Issues Pull Requests 11 Packages Projects Releases 10 Wiki Activity

docs(skills): localize research skill docs

This commit is contained in:
NeeNe 2026-07-05 13:30:27 +08:00
parent 8642f4026e
commit 5eb7768f0e
17 changed files with 176 additions and 190 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

24
doc/changes/research-skills-suite.md
View File

@ -1,22 +1,22 @@
# Research skills suite
# 科研 Skills 套件
Adds seven read-only research-oriented Agent Skills for GitLink repositories:
本次新增 7 个面向 GitLink 科研仓库的只读 Agent Skills：
- `gitlink-research-reproducibility`: reproducibility checklist and remediation plan for research code.
- `gitlink-research-compliance`: license, security policy, dependency, and sensitive-file risk audit.
- `gitlink-research-progress-tracker`: weekly progress report and risk early warning for research projects.
- `gitlink-research-collaboration-map`: contributor, Issue, and PR collaboration mapping for research teams.
- `gitlink-research-knowledge-graph`: keyword-based research landscape and lightweight knowledge graph workflow.
- `gitlink-research-data-provenance`: dataset source, citation, license, and privacy-risk audit.
- `gitlink-research-artifact-handbook`: research artifact handbook generation for demos, handoff, and open-source release.
- `gitlink-research-reproducibility`：科研代码复现性清单和整改计划。
- `gitlink-research-compliance`：许可证、安全策略、依赖和敏感文件风险审计。
- `gitlink-research-progress-tracker`：科研项目周报和风险预警。
- `gitlink-research-collaboration-map`：面向课题组的贡献者、Issue 和 PR 协作画像。
- `gitlink-research-knowledge-graph`：基于关键词的科研生态调研和轻量知识图谱工作流。
- `gitlink-research-data-provenance`：数据集来源、引用、许可证和隐私风险审计。
- `gitlink-research-artifact-handbook`：面向答辩、交接和开源发布的科研成果手册生成。
These Skills are designed as standalone `gitlink-cli` Skill PR deliverables. They do not depend on external project code and can also be composed by end-to-end research workflows.
这些 Skills 按独立 `gitlink-cli` Skill PR 交付物设计，不依赖外部项目代码，也可被端到端科研工作流组合调用。
Validation:
验证方式：
```bash
make validate-research-skills
git diff --check
```
Agent verification guidance is documented in `docs/research-skills-agent-test-report.md`.
真实 Agent 验证说明见 `docs/research-skills-agent-test-report.md`。

144
docs/research-skills-agent-test-report.md
View File

@ -1,8 +1,8 @@
# Research Skills Agent Test Report
# 科研 Skills Agent 测试报告
## Scope
## 覆盖范围
This report covers the seven read-only research Skills added for the GitLink competition:
本报告覆盖本次为 GitLink 竞赛新增的 7 个只读科研 Skills：
- `gitlink-research-reproducibility`
- `gitlink-research-compliance`
@ -12,136 +12,136 @@ This report covers the seven read-only research Skills added for the GitLink com
- `gitlink-research-data-provenance`
- `gitlink-research-artifact-handbook`
The Skills are standalone `gitlink-cli` Agent Skills. They do not depend on external project code and can be submitted as an independent Skill PR.
这些 Skills 均为独立的 `gitlink-cli` Agent Skills，不依赖外部项目代码，可作为独立 Skill PR 提交。
## Automatic Validation
## 自动验证
Run:
执行：
```bash
make validate-research-skills
git diff --check
```
The validation script checks:
验证脚本检查以下内容：
- every Skill has valid `SKILL.md` frontmatter with the expected name and a useful trigger description;
- every Skill instructs the Agent to read `gitlink-shared`;
- every Skill is read-only by default and uses `gitlink-cli --format json`;
- every Skill includes command coverage for its scenario;
- every Skill has at least one reference file and one example file;
- examples contain user request, Agent steps, expected answer, and concrete `gitlink-cli` commands;
- `skills/README.md` and `doc/changes/research-skills-suite.md` index all seven Skills.
- 每个 Skill 的 `SKILL.md` 均具备合法 frontmatter、预期名称和可触发描述；
- 每个 Skill 均要求 Agent 先读取 `gitlink-shared`；
- 每个 Skill 默认只读，并要求使用 `gitlink-cli --format json`；
- 每个 Skill 覆盖对应场景所需命令；
- 每个 Skill 至少包含一个参考文件和一个示例文件；
- 示例包含用户请求、Agent 步骤、预期回答和具体 `gitlink-cli` 命令；
- `skills/README.md` 和 `doc/changes/research-skills-suite.md` 已索引全部 7 个 Skills。
## Offline Agent Test Prompts
## 离线 Agent 测试提示词
These prompts can be used in Claude Code, Cursor, Codex, or OpenClaw after installing the Skills.
安装 Skills 后，可在 Claude Code、Cursor、Codex 或 OpenClaw 中使用以下提示词测试。
### Reproducibility
### 复现性审计
```text
Use gitlink-research-reproducibility to audit songhui18/ICCV2021. Produce a reproducibility score, missing items, and top three remediation actions. Do not write back to GitLink.
请使用 gitlink-research-reproducibility 审计 songhui18/ICCV2021，输出复现性评分、缺失项和前三个整改动作。不要向 GitLink 写回任何内容。
```
Expected behavior:
预期行为：
- read `gitlink-shared`;
- run `repo +info` and `repo +tree`, or fall back to `api GET /sub_entries`;
- cite command sources;
- produce a checklist report.
- 读取 `gitlink-shared`；
- 执行 `repo +info` 和 `repo +tree`，必要时回退到 `api GET /sub_entries`；
- 标注命令数据来源；
- 输出清单式报告。
### Compliance
### 合规预审
```text
Use gitlink-research-compliance to check Gitconomy/Git4Research before open-source release. Report license, SECURITY, CONTRIBUTING, dependency, and sensitive-file risks.
请使用 gitlink-research-compliance 检查 Gitconomy/Git4Research 的开源发布准备情况，报告许可证、SECURITY、CONTRIBUTING、依赖和敏感文件风险。
```
Expected behavior:
预期行为：
- perform read-only file tree checks;
- avoid printing sensitive contents;
- state that the output is an engineering pre-check, not legal advice.
- 执行只读文件树检查；
- 避免打印敏感内容；
- 说明输出是工程预检查，不构成法律意见。
### Progress Tracking
### 进度跟踪
```text
Use gitlink-research-progress-tracker to generate a weekly research progress report for songhui18/ICCV2021, including Issue/PR queues and risk warnings.
请使用 gitlink-research-progress-tracker 为 songhui18/ICCV2021 生成科研项目周报，包含 Issue/PR 队列和风险预警。
```
Expected behavior:
预期行为：
- collect repo, Issue, PR, Release, and contributor data;
- separate facts from recommendations;
- avoid posting a comment unless explicitly requested.
- 采集仓库、Issue、PR、Release 和贡献者数据；
- 区分事实和建议；
- 除非用户明确要求，不发布评论。
### Collaboration Map
### 协作画像
```text
Use gitlink-research-collaboration-map to analyze collaboration structure for Gitconomy/Git4Research. Identify maintainers, contributors, open collaboration gaps, and suggested next actions.
请使用 gitlink-research-collaboration-map 分析 Gitconomy/Git4Research 的协作结构，识别维护者、贡献者、协作缺口和下一步建议。
```
Expected behavior:
预期行为：
- use public contributor, Issue, PR, language, and optional user data;
- avoid private contact information;
- base recommendations on evidence.
- 使用公开贡献者、Issue、PR、语言和可选用户数据；
- 避免输出私人联系方式；
- 建议必须有证据支撑。
### Knowledge Graph
### 知识图谱
```text
Use gitlink-research-knowledge-graph to search GitLink for “论文复现” and “open research”, then output a Markdown trend summary and graph JSON.
请使用 gitlink-research-knowledge-graph 在 GitLink 搜索“论文复现”和“open research”，输出 Markdown 趋势摘要和图谱 JSON。
```
Expected behavior:
预期行为：
- run bounded keyword searches;
- deep-analyze no more than eight repositories;
- produce `nodes` and `edges` using the documented graph schema;
- mark platform-data limits.
- 执行有边界的关键词搜索；
- 深度分析不超过 8 个仓库；
- 按图谱 schema 输出 `nodes` 和 `edges`；
- 标注平台数据局限。
### Data Provenance
### 数据来源审计
```text
Use gitlink-research-data-provenance to audit songhui18/ICCV2021 for dataset source, citation, license, and privacy risks. Do not print raw data contents.
请使用 gitlink-research-data-provenance 审计 songhui18/ICCV2021 的数据集来源、引用、许可证和隐私风险。不要打印原始数据内容。
```
Expected behavior:
预期行为：
- collect repo info and selected file trees;
- report paths and risk types only;
- distinguish missing evidence from confirmed risk.
- 采集仓库信息和选定文件树；
- 只报告路径和风险类型；
- 区分证据缺失和已确认风险。
### Artifact Handbook
### 成果手册
```text
Use gitlink-research-artifact-handbook to generate a research artifact handbook for Gitconomy/Git4Research, suitable for project handoff and competition demo.
请使用 gitlink-research-artifact-handbook 为 Gitconomy/Git4Research 生成科研成果沉淀手册，用于项目交接和竞赛演示。
```
Expected behavior:
预期行为：
- collect repo, README, tree, Issue/PR, Release, language, and contributor signals when available;
- produce a structured handbook with missing fields marked as `待补充`;
- avoid writing Wiki, Issue, Release, or repository files unless explicitly requested.
- 在可用时采集仓库、README、文件树、Issue/PR、Release、语言和贡献者信号；
- 输出结构化手册，缺失字段标注为 `待补充`；
- 除非用户明确要求，不写入 Wiki、Issue、Release 或仓库文件。
## Real Agent Verification
## 真实 Agent 验证
For competition submission, record at least one Agent run with screenshots or video:
竞赛提交时，建议至少录制一次 Agent 运行过程截图或视频：
1. Install or expose the Skills to the Agent.
2. Run one prompt from this report against a public GitLink repository.
3. Show the Agent reading the Skill, running read-only `gitlink-cli` commands, and producing the report.
4. Show that no write operation was performed.
1. 安装或暴露这些 Skills 给 Agent。
2. 选择一个公开 GitLink 仓库运行本报告中的提示词。
3. 展示 Agent 读取 Skill、执行只读 `gitlink-cli` 命令并生成报告。
4. 展示未执行任何写操作。
Recommended demo target:
推荐演示目标：
```text
songhui18/ICCV2021
```
This repository has richer Issue and PR data, which makes progress tracking and collaboration mapping more visible.
该仓库具备较丰富的 Issue 和 PR 数据，便于展示进度跟踪和协作画像。
## Known Limitations
## 已知限制
- The automatic script validates Skill structure and workflow instructions; it does not prove semantic quality of an LLM-generated report.
- Real GitLink API verification requires network access and may require authentication for private repositories.
- Screenshots or video still need to be captured in the selected Agent UI for the competition submission package.
- 自动脚本验证 Skill 结构和工作流说明，不证明大模型生成报告的语义质量。
- 真实 GitLink API 验证需要网络访问；私有仓库可能需要认证。
- 竞赛提交材料仍需在选定 Agent 界面中补充截图或视频。

28
scripts/validate-research-skills.py
View File

@ -61,20 +61,20 @@ def read(path: Path) -> str:
try:
return path.read_text(encoding="utf-8")
except FileNotFoundError:
fail(f"missing file: {path}")
fail(f"缺少文件：{path}")
def parse_frontmatter(text: str, path: Path) -> dict[str, str]:
if not text.startswith("---\n"):
fail(f"{path} does not start with YAML frontmatter")
fail(f"{path} 未以 YAML frontmatter 开头")
try:
_, frontmatter, _ = text.split("---", 2)
except ValueError:
fail(f"{path} has incomplete YAML frontmatter")
fail(f"{path} 的 YAML frontmatter 不完整")
parsed: dict[str, str] = {}
for line in frontmatter.strip().splitlines():
if ":" not in line:
fail(f"{path} frontmatter line has no colon: {line}")
fail(f"{path} 的 frontmatter 行缺少冒号：{line}")
key, value = line.split(":", 1)
parsed[key.strip()] = value.strip().strip('"')
return parsed
@ -82,12 +82,12 @@ def parse_frontmatter(text: str, path: Path) -> dict[str, str]:
def assert_contains(text: str, needle: str, path: Path) -> None:
if needle not in text:
fail(f"{path} is missing required text: {needle}")
fail(f"{path} 缺少必需文本：{needle}")
def assert_regex(text: str, pattern: str, path: Path) -> None:
if not re.search(pattern, text, flags=re.MULTILINE):
fail(f"{path} does not match required pattern: {pattern}")
fail(f"{path} 未匹配必需模式：{pattern}")
def validate_skill(root: Path, name: str, spec: dict[str, list[str]]) -> None:
@ -97,13 +97,13 @@ def validate_skill(root: Path, name: str, spec: dict[str, list[str]]) -> None:
frontmatter = parse_frontmatter(text, skill_md)
if frontmatter.get("name") != name:
fail(f"{skill_md} has wrong name: {frontmatter.get('name')}")
fail(f"{skill_md} 的 name 字段错误：{frontmatter.get('name')}")
description = frontmatter.get("description", "")
if len(description) < 30:
fail(f"{skill_md} description is too short")
fail(f"{skill_md} 的 description 过短")
for keyword in spec["keywords"]:
if keyword not in text and keyword not in description:
fail(f"{skill_md} is missing scenario keyword: {keyword}")
fail(f"{skill_md} 缺少场景关键词：{keyword}")
assert_contains(text, "gitlink-shared", skill_md)
assert_contains(text, "gitlink-cli", skill_md)
@ -118,14 +118,14 @@ def validate_skill(root: Path, name: str, spec: dict[str, list[str]]) -> None:
path = skill_dir / "references" / filename
ref_text = read(path)
if len(ref_text.strip().splitlines()) < 5:
fail(f"{path} is too small to be useful")
fail(f"{path} 内容过少，无法作为有效参考")
for filename in spec["examples"]:
path = skill_dir / "examples" / filename
example_text = read(path)
assert_contains(example_text, "User request", path)
assert_contains(example_text, "Agent steps", path)
assert_contains(example_text, "Expected answer", path)
assert_contains(example_text, "用户请求", path)
assert_contains(example_text, "Agent 步骤", path)
assert_contains(example_text, "预期回答", path)
assert_contains(example_text, "gitlink-cli", path)
@ -142,7 +142,7 @@ def validate_change_note(root: Path) -> None:
for name in RESEARCH_SKILLS:
assert_contains(text, name, note)
if "scripts/validate-research-skills.py" not in text and "make validate-research-skills" not in text:
fail(f"{note} is missing the research Skills validation command")
fail(f"{note} 缺少科研 Skills 验证命令")
def main() -> int:

15
skills/gitlink-research-artifact-handbook/examples/artifact-handbook.md
View File

@ -1,12 +1,12 @@
# Artifact handbook example
# 成果手册示例
User request:
用户请求：
```text
请为 Gitconomy/Git4Research 生成一份科研成果沉淀手册，用于课题组交接和答辩展示。
```
Agent steps:
Agent 步骤：
```bash
gitlink-cli repo +info --owner Gitconomy --repo Git4Research --format json
@ -17,9 +17,8 @@ gitlink-cli pr +list --owner Gitconomy --repo Git4Research --state merged --form
gitlink-cli release +list --owner Gitconomy --repo Git4Research --format json
```
Expected answer:
- Produce a handbook using the template.
- Mark missing evidence as `待补充`.
- Confirm that no Wiki, Issue, PR, Release, or file write operation was performed.
预期回答：
- 按模板生成科研成果沉淀手册。
- 将缺失证据标注为 `待补充`。
- 确认未执行 Wiki、Issue、PR、Release 或文件写操作。

3
skills/gitlink-research-artifact-handbook/references/handbook-template.md
View File

@ -1,4 +1,4 @@
# Research artifact handbook template
# 科研成果沉淀手册模板
```markdown
# 科研成果沉淀手册：<owner>/<repo>
@ -56,4 +56,3 @@
- [ ] 实验命令可运行
- [ ] Release 或归档版本可引用
```

15
skills/gitlink-research-collaboration-map/examples/collaboration-map.md
View File

@ -1,12 +1,12 @@
# Collaboration map example
# 协作画像示例
User request:
用户请求：
```text
请分析 Gitconomy/Git4Research 的科研协作结构，找出核心维护者和可跟进的问题方向。
```
Agent steps:
Agent 步骤：
```bash
gitlink-cli repo +info --owner Gitconomy --repo Git4Research --format json
@ -15,9 +15,8 @@ gitlink-cli issue +list --owner Gitconomy --repo Git4Research --state open --for
gitlink-cli pr +list --owner Gitconomy --repo Git4Research --state open --format json
```
Expected answer:
- Separate facts from recommendations.
- Mention data limits if the repository has little Issue/PR activity.
- Do not expose private contact information.
预期回答：
- 区分事实和建议。
- 如果仓库 Issue/PR 活动较少，需要说明数据局限。
- 不暴露私人联系方式。

21
skills/gitlink-research-collaboration-map/references/profile-fields.md
View File

@ -1,20 +1,19 @@
# Collaboration profile fields
# 协作画像字段
| 字段 | 来源 | 用途 |
|---|---|---|
| contributor login/name | `repo +contributors` | 识别代码贡献者 |
| commits/contributions | `repo +contributors` | 粗略衡量贡献量 |
| issue author | `issue +list` | 识别问题提出者 |
| issue assignee | `issue +list` | 识别责任分配 |
| issue labels | `issue +list` | 判断研究主题或任务类型 |
| PR author | `pr +list` | 识别代码贡献者和集成者 |
| PR status | `pr +list` | 判断协作流转状态 |
| language | `repo +languages` | 推断技术栈 |
| 贡献者账号/名称 | `repo +contributors` | 识别代码贡献者 |
| 提交/贡献计数 | `repo +contributors` | 粗略衡量贡献量 |
| Issue 作者 | `issue +list` | 识别问题提出者 |
| Issue 负责人 | `issue +list` | 识别责任分配 |
| Issue 标签 | `issue +list` | 判断研究主题或任务类型 |
| PR 作者 | `pr +list` | 识别代码贡献者和集成者 |
| PR 状态 | `pr +list` | 判断协作流转状态 |
| 语言 | `repo +languages` | 推断技术栈 |
Suggested role labels:
建议角色标签：
- 核心维护者：代码贡献和 Issue/PR 参与均较多。
- 问题提出者：Issue 活跃但代码贡献少，适合需求澄清和复现实验。
- 实验贡献者：PR 或实验相关 Issue 活跃。
- 潜在协作者：与主要语言或研究主题匹配，但当前参与较少。

15
skills/gitlink-research-compliance/examples/compliance-audit.md
View File

@ -1,21 +1,20 @@
# Compliance audit example
# 合规预审示例
User request:
用户请求：
```text
请检查 Gitconomy/Git4Research 是否具备开源发布合规基础。
```
Agent steps:
Agent 步骤：
```bash
gitlink-cli repo +info --owner Gitconomy --repo Git4Research --format json
gitlink-cli repo +tree --owner Gitconomy --repo Git4Research --ref main --format json
```
Expected answer:
- Report license, security policy, contribution guide, dependency manifest, and sensitive file risks.
- Mention whether the audit is root-only or includes selected subdirectories.
- Avoid quoting or exposing sensitive file contents.
预期回答：
- 报告许可证、安全策略、贡献指南、依赖清单和敏感文件风险。
- 说明审计范围是仅根目录，还是包含选定子目录。
- 避免引用或暴露敏感文件内容。

3
skills/gitlink-research-compliance/references/risk-rules.md
View File

@ -1,4 +1,4 @@
# Compliance risk rules
# 合规风险规则
| 风险 | 等级 | 判定 | 建议 |
|---|---|---|---|
@ -8,4 +8,3 @@
| 缺少依赖声明 | 中 | 无常见依赖清单 | 补充依赖文件，便于许可证和漏洞审计 |
| 缺少 SECURITY | 中 | 无 `SECURITY.md` | 补充漏洞披露流程 |
| 缺少 CONTRIBUTING | 低 | 无 `CONTRIBUTING.md` | 补充贡献流程和代码规范 |

15
skills/gitlink-research-data-provenance/examples/data-provenance-audit.md
View File

@ -1,12 +1,12 @@
# Data provenance audit example
# 数据来源审计示例
User request:
用户请求：
```text
请检查 songhui18/ICCV2021 是否有科研数据来源、数据许可证或隐私风险问题。
```
Agent steps:
Agent 步骤：
```bash
gitlink-cli repo +info --owner songhui18 --repo ICCV2021 --format json
@ -14,9 +14,8 @@ gitlink-cli repo +tree --owner songhui18 --repo ICCV2021 --ref master --format j
gitlink-cli repo +tree --owner songhui18 --repo ICCV2021 --path data --ref master --format json
```
Expected answer:
- Report only paths and risk types, not raw data contents.
- Separate data source, license, citation, privacy, and large-file risks.
- State whether the assessment is root-only or includes selected subdirectories.
预期回答：
- 只报告路径和风险类型，不输出原始数据内容。
- 区分数据来源、许可证、引用、隐私和大文件风险。
- 说明评估范围是仅根目录，还是包含选定子目录。

3
skills/gitlink-research-data-provenance/references/provenance-rules.md
View File

@ -1,4 +1,4 @@
# Data provenance rules
# 数据来源规则
| 检查项 | 风险等级 | 判定线索 | 建议 |
|---|---|---|---|
@ -9,4 +9,3 @@
| 数据下载脚本无校验 | 中 | 有 download/get_data 脚本但无 checksum/version | 补充 checksum、版本号和失败处理 |
| 引用链不完整 | 中 | README 未说明数据、论文、代码之间的关系 | 补充数据引用、论文引用和实验脚本映射 |
| 未发现数据痕迹 | 低 | 未见数据目录、下载脚本或数据扩展名 | 在报告中标注未发现，建议说明数据是否需另行申请 |

15
skills/gitlink-research-knowledge-graph/examples/keyword-map.md
View File

@ -1,12 +1,12 @@
# Keyword knowledge graph example
# 关键词知识图谱示例
User request:
用户请求：
```text
请围绕“论文复现”和“open research”在 GitLink 上做科研热点搜索，并输出轻量知识图谱。
```
Agent steps:
Agent 步骤：
```bash
gitlink-cli search +repos -k "论文复现" --format json
@ -16,9 +16,8 @@ gitlink-cli repo +languages --owner <owner> --repo <repo> --format json
gitlink-cli repo +contributors --owner <owner> --repo <repo> --format json
```
Expected answer:
- List keywords and candidate repositories.
- Provide a Markdown trend report.
- Provide `nodes` and `edges` JSON using `references/graph-schema.md`.
预期回答：
- 列出关键词和候选仓库。
- 提供 Markdown 趋势报告。
- 按 `references/graph-schema.md` 输出包含 `nodes` 和 `edges` 的 JSON。

27
skills/gitlink-research-knowledge-graph/references/graph-schema.md
View File

@ -1,27 +1,27 @@
# Research knowledge graph schema
# 科研知识图谱结构
Node types:
节点类型：
| Type | Required fields | Description |
| 类型 | 必填字段 | 说明 |
|---|---|---|
| `repo` | `id`, `owner`, `repo`, `label` | GitLink repository |
| `language` | `id`, `name` | Programming language |
| `contributor` | `id`, `login` | Public contributor identity |
| `topic` | `id`, `name` | Keyword or inferred research topic |
| `issue` | `id`, `title`, `state` | Representative Issue |
| `pr` | `id`, `title`, `state` | Representative PR |
| `repo` | `id`, `owner`, `repo`, `label` | GitLink 仓库 |
| `language` | `id`, `name` | 编程语言 |
| `contributor` | `id`, `login` | 公开贡献者身份 |
| `topic` | `id`, `name` | 关键词或推断出的研究主题 |
| `issue` | `id`, `title`, `state` | 代表性 Issue |
| `pr` | `id`, `title`, `state` | 代表性 PR |
Edge types:
边类型：
| Type | From | To | Evidence |
| 类型 | 起点 | 终点 | 证据 |
|---|---|---|---|
| `uses_language` | repo | language | `repo +languages` |
| `contributed_by` | contributor | repo | `repo +contributors` |
| `matches_topic` | repo | topic | search keyword or description |
| `matches_topic` | repo | topic | 搜索关键词或描述 |
| `has_issue` | repo | issue | `issue +list` |
| `has_pr` | repo | pr | `pr +list` |
Example:
示例：
```json
{
@ -34,4 +34,3 @@ Example:
]
}
```

15
skills/gitlink-research-progress-tracker/examples/weekly-report.md
View File

@ -1,12 +1,12 @@
# Research weekly report example
# 科研周报示例
User request:
用户请求：
```text
请给 songhui18/ICCV2021 生成一份科研项目进度周报和风险预警。
```
Agent steps:
Agent 步骤：
```bash
gitlink-cli repo +info --owner songhui18 --repo ICCV2021 --format json
@ -17,9 +17,8 @@ gitlink-cli pr +list --owner songhui18 --repo ICCV2021 --state merged --format j
gitlink-cli release +list --owner songhui18 --repo ICCV2021 --format json
```
Expected answer:
- Provide a concise weekly report.
- Identify the top 3 risks with evidence.
- Do not write comments back unless explicitly requested.
预期回答：
- 提供简洁的科研项目周报。
- 识别前三个风险并给出证据。
- 除非用户明确要求，不写回评论。

3
skills/gitlink-research-progress-tracker/references/risk-model.md
View File

@ -1,4 +1,4 @@
# Research progress risk model
# 科研进度风险模型
| 风险 | 等级 | 判定线索 | 建议 |
|---|---|---|---|
@ -8,4 +8,3 @@
| 维护停滞 | 高 | 仓库更新时间较久且仍有开放问题 | 明确是否归档、继续维护或招募维护者 |
| 单点维护 | 中 | 贡献者数量很少 | 补贡献指南和 onboarding 任务 |
| 数据不足 | 低 | API 缺失或仓库未启用对应功能 | 标注数据不可用，不推断过度结论 |

17
skills/gitlink-research-reproducibility/examples/reproducibility-audit.md
View File

@ -1,12 +1,12 @@
# Reproducibility audit example
# 复现性审计示例
User request:
用户请求：
```text
请用 GitLink 分析 songhui18/ICCV2021 的论文复现性，输出缺失项和整改优先级。
```
Agent steps:
Agent 步骤：
```bash
gitlink-cli repo +info --owner songhui18 --repo ICCV2021 --format json
@ -14,10 +14,9 @@ gitlink-cli repo +tree --owner songhui18 --repo ICCV2021 --ref master --format j
gitlink-cli api GET /songhui18/ICCV2021/sub_entries --query 'filepath=&ref=master' --format json
```
Expected answer:
- Quote the collected command names.
- Score README, LICENSE, dependencies, tests, CI, examples/docs, data notes.
- List the top three fixes before any optional polishing.
- State that no write action was performed.
预期回答：
- 列出已采集的命令名称。
- 对 README、LICENSE、依赖、测试、CI、示例/文档和数据说明评分。
- 优先列出前三个整改项，再给可选优化建议。
- 声明未执行写操作。

3
skills/gitlink-research-reproducibility/references/checklist.md
View File

@ -1,4 +1,4 @@
# Reproducibility checklist
# 复现性检查清单
| 项目 | 分值 | 通过条件 | 缺失时建议 |
|---|---:|---|---|
@ -9,4 +9,3 @@
| CI 配置 | 10 | 存在 CI 配置目录或文件 | 增加基础 lint/test CI |
| 示例或文档 | 10 | 存在 docs/examples/demo | 补充复现实验步骤或示例输入输出 |
| 数据或实验说明 | 10 | 存在 data/datasets/experiments/notebooks 或 README 中明确说明 | 写明数据来源、下载方式、实验配置 |