Token导航 LogoToken导航

Agent Skill · 研究检索

github-evidence-kit(GitHub evidence KIT 搜索)

用于围绕 GitHub 仓库、Issue、Pull Request、分支、提交和代码协作流程提供辅助能力。它适合让 Agent 查询项目状态、整理变更、辅助创建或检查协作事项,并把仓库中的信息转成可执行的下一步。使用时需要区分只读查询和写入操作;涉及创建 PR、修改 Issue、推送分支或访问私有仓库时,应确认 token 权限、目标仓库范围和用户授权。

github-evidence-kit 是研究检索类 Agent Skill,可用宿主为 Codex、Claude、Cursor、Gemini CLI,通过GitHub 仓库获取,权限为需要联网;许可证为 MIT,第三方扫描有提醒项,按安装量在研究检索类的 73,675 个 Skill 里排第 56,831。

总安装 445周安装 18GitHub Stars 1,974下载 140核验 2026-05-01

安装说明

本站只整理中文说明,不托管安装包,也不代用户安装。
GitHub
许可证
MIT

安装方式

通过对话安装

复制提示词发给支持本地命令或 Skills 的 AI 助手,先确认命令和权限,再让它执行。

请帮我安装这个 Agent Skill:github-evidence-kit(GitHub evidence KIT 搜索)
来源仓库:https://github.com/gadievron/raptor
仓库路径:skills/github-evidence-kit
安装命令:
npx skills add https://github.com/gadievron/raptor --skill github-evidence-kit
安装前请先检查当前环境是否支持对应 CLI,并向我确认将要执行的命令、安装目录、联网范围和文件读写权限;确认后再执行。

命令行安装

复制命令到本机终端执行。不同来源提供的安装方式可能略有差异;本站展示可直接复制的安装命令,安装前请核对来源页面。

skills.shnpx skills
npx skills add https://github.com/gadievron/raptor --skill github-evidence-kit

这个 Skill 是做什么的

研究检索类 Skill 主要解决「找资料、核对事实、把来源整理清楚」的问题,通常会让助手先检索再总结。

收集 GitHub 活动作为证据链,支撑审计或合规报告。适用于导出提交记录、PR 评审日志与 issue 解决时间线。生成标准化 JSON 或 PDF 文档,符合法律取证格式要求。敏感操作日志需加密存储,防止未授权人员查看原始数据。

github-evidence-kit 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

来源仓库内路径:skills/github-evidence-kit

上手步骤

4 步
  1. 1

    确认自己的宿主工具

    公开资料标注支持 Codex、Claude、Cursor、Gemini CLI;先确认自己用的是其中之一,再按对应工具的 Skills 目录放置。

  2. 2

    拿到安装命令

    可以直接复制的命令是 npx skills add https://github.com/gadievron/raptor --skill github-evidence-kit;命令里的占位符换成自己的路径或密钥再执行。

  3. 3

    安装后重启宿主

    命令执行完要完全退出宿主工具再打开,让新装的 Skill 被重新扫描到,否则列表里可能看不到。

  4. 4

    按最小权限试用

    它需要需要联网,先在测试目录或测试账号里跑,确认没问题再放到正式项目。

常见问题

6 条

github-evidence-kit 怎么安装?

在终端执行 npx skills add https://github.com/gadievron/raptor --skill github-evidence-kit。装完完全退出宿主工具再打开,让它重新扫描本地 Skill。

github-evidence-kit 支持哪些 AI 工具?

公开资料标注支持 Codex、Claude、Cursor、Gemini CLI。其他工具只要能读取本地 Skill 目录,一般也能手动放进去,但要自己确认触发方式是否一致。

github-evidence-kit 需要什么权限?

权限级别是需要联网。该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

github-evidence-kit 是免费的吗?

来源标注的许可证是 MIT。本站不代收任何费用,也不托管安装包;能否商用、要不要署名,以许可证原文为准。

github-evidence-kit 用的人多吗?

公开数据是累计安装 445 次,近一周安装 18 次,下载 140 次,GitHub 1,974 星。这些数字来自第三方目录统计,只能当作热度参考,不代表质量或安全性。

github-evidence-kit 安全吗?

第三方扫描有提醒项,说明存在需要人工确认的行为。安装前请打开源码看清它读了什么、联网到哪里。

值得选的地方和要先想清楚的地方

值得选的地方

  • 同时适配 4 种工具,换客户端不用重新找替代品。
  • 3 个目录同时收录,说明这个 Skill 流传比较广。
  • 许可证明确为 MIT,商用和分发边界好判断。
  • 来源仓库有 1,974 星,维护活跃度相对可查。

先想清楚的地方

  • 第三方扫描存在提醒项或未通过项,需要人工逐条确认。
  • 第三方目录整理,版本更新和行为细节以来源仓库为准。

适不适合你

更适合

  • 已经在用 Codex,想直接扩展它能力的人。
  • 需要处理研究检索类任务、又不想每次都手写提示词的人。

不太适合

  • 需要厂商级技术支持和 SLA 的生产系统。

研究检索类 Skill 对比

6 个同类
Skill总安装GitHub 星权限许可证扫描
find-skills

寻找技能 · Codex、Claude

31,199,81316,486操作浏览器MIT有提醒
self-improving-agent

自我改进 Agent · OpenClaw

16,464,0023,396接触敏感数据MIT-0通过
soultrace

灵魂追踪 · Codex、Claude

7,588,94412需要联网未明确有提醒
self-improving

自我完善 · OpenClaw

6,829,2841,046需要联网MIT-0通过
microsoft-foundry

微软代工厂 · Codex、Claude

6,614,144764调用外部服务未明确有提醒
github-evidence-kit

GitHub evidence KIT 搜索 · 本页

4451,974需要联网MIT有提醒

按安装量取同分类前几名;数字来自第三方目录统计,只作热度参考,不代表质量结论。

SKILL.md 原文

GH Evidence Kit

Purpose: Create, store, and verify forensic evidence from GitHub-related public sources and local git repositories.

When to Use This Skill

  • Creating verifiable evidence objects from GitHub activity
  • Local git forensics - analyzing cloned repositories, dangling commits, reflog
  • Exporting evidence collections to JSON for sharing/archival
  • Loading and re-verifying previously collected evidence
  • Recovering deleted GitHub content (issues, PRs, commits) from GH Archive
  • Tracking IOCs (Indicators of Compromise) with source verification

Quick Start

from src.collectors import GitHubAPICollector, LocalGitCollector, GHArchiveCollector
from src import EvidenceStore

# Create collectors for different sources
github = GitHubAPICollector()
local = LocalGitCollector("/path/to/repo")
archive = GHArchiveCollector()

# Collect evidence from GitHub API
commit = github.collect_commit("aws", "aws-toolkit-vscode", "678851b...")
pr = github.collect_pull_request("aws", "aws-toolkit-vscode", 7710)

# Collect evidence from local git (first-class forensic source)
local_commit = local.collect_commit("HEAD")
dangling = local.collect_dangling_commits()  # Forensic gold!

# Store and export
store = EvidenceStore()
store.add(commit)
store.add(pr)
store.add(local_commit)
store.add_all(dangling)
store.save("evidence.json")

# Verify all evidence against original sources
is_valid, errors = store.verify_all()

Collectors

GitHubAPICollector

Collects evidence from the live GitHub API.

from src.collectors import GitHubAPICollector

collector = GitHubAPICollector()
MethodReturns
collect_commit(owner, repo, sha)CommitObservation
collect_issue(owner, repo, number)IssueObservation
collect_pull_request(owner, repo, number)IssueObservation
collect_file(owner, repo, path, ref)FileObservation
collect_branch(owner, repo, branch_name)BranchObservation
collect_tag(owner, repo, tag_name)TagObservation
collect_release(owner, repo, tag_name)ReleaseObservation
collect_forks(owner, repo)list[ForkObservation]

LocalGitCollector (First-Class Forensics)

Collects evidence from local git repositories. Essential for forensic analysis of cloned repos.

from src.collectors import LocalGitCollector

collector = LocalGitCollector("/path/to/cloned/repo")

# Collect a specific commit
commit = collector.collect_commit("HEAD")
commit = collector.collect_commit("abc123")

# Find dangling commits (not reachable from any ref)
# This is forensic gold - reveals force-pushed or deleted commits!
dangling = collector.collect_dangling_commits()
for commit in dangling:
    print(f"Found dangling: {commit.sha[:8]} - {commit.message}")
MethodReturns
collect_commit(sha)CommitObservation
collect_dangling_commits()list[CommitObservation]

GHArchiveCollector

Collects and recovers evidence from GH Archive (BigQuery). Requires credentials.

from src.collectors import GHArchiveCollector

collector = GHArchiveCollector()

# Query events by timestamp (YYYYMMDDHHMM format)
events = collector.collect_events(
    timestamp="202507132037",
    repo="aws/aws-toolkit-vscode"
)

# Recover deleted content
deleted_issue = collector.recover_issue("aws/aws-toolkit-vscode", 123, "2025-07-13T20:30:24Z")
deleted_pr = collector.recover_pr("aws/aws-toolkit-vscode", 7710, "2025-07-13T20:30:24Z")
deleted_commit = collector.recover_commit("aws/aws-toolkit-vscode", "678851b", "2025-07-13T20:30:24Z")
force_pushed = collector.recover_force_push("aws/aws-toolkit-vscode", "2025-07-13T20:30:24Z")
MethodReturns
collect_events(timestamp, repo, actor, event_type)list[Event]
recover_issue(repo, number, timestamp)IssueObservation
recover_pr(repo, number, timestamp)IssueObservation
recover_commit(repo, sha, timestamp)CommitObservation
recover_force_push(repo, timestamp)CommitObservation

WaybackCollector

Collects archived snapshots from the Wayback Machine.

from src.collectors import WaybackCollector

collector = WaybackCollector()

# Get all snapshots for a URL
snapshots = collector.collect_snapshots("https://github.com/owner/repo")

# With date filtering
snapshots = collector.collect_snapshots(
    "https://github.com/owner/repo",
    from_date="20250101",
    to_date="20250731"
)

# Fetch actual content of a snapshot
content = collector.collect_snapshot_content(
    "https://github.com/owner/repo",
    "20250713203024"  # YYYYMMDDHHMMSS format
)

Verification

Verification is separated from data collection. Use ConsistencyVerifier to validate evidence against original sources.

from src.verifiers import ConsistencyVerifier

verifier = ConsistencyVerifier()

# Verify single evidence
result = verifier.verify(commit)
if not result.is_valid:
    print(f"Errors: {result.errors}")

# Verify multiple
result = verifier.verify_all([commit, pr, issue])

Or use the convenience method on EvidenceStore:

store = EvidenceStore()
store.add_all([commit, pr, issue])
is_valid, errors = store.verify_all()

EvidenceStore

Store, query, and export evidence collections.

from src import EvidenceStore
from datetime import datetime

store = EvidenceStore()

# Add evidence
store.add(commit)
store.add_all([pr, issue, ioc])

# Query
commits = store.filter(observation_type="commit")
recent = store.filter(after=datetime(2025, 7, 1))
from_github = store.filter(source="github")
from_git = store.filter(source="git")
repo_events = store.filter(repo="aws/aws-toolkit-vscode")

# Export/Import
store.save("evidence.json")
store = EvidenceStore.load("evidence.json")

# Summary
print(store.summary())
# {'total': 5, 'events': {...}, 'observations': {...}, 'by_source': {...}}

# Verify all against sources
is_valid, errors = store.verify_all()

Loading Evidence from JSON

from src import load_evidence_from_json
import json

with open("evidence.json") as f:
    data = json.load(f)

for item in data:
    evidence = load_evidence_from_json(item)
    # Evidence is now a typed Pydantic model

Evidence Types

Events (from GH Archive)

All 12 GitHub event types are supported:

TypeDescription
PushEventCommits pushed
PullRequestEventPR opened/closed/merged
IssueEventIssue opened/closed
IssueCommentEventComment on issue/PR
CreateEventBranch/tag created
DeleteEventBranch/tag deleted
ForkEventRepository forked
WatchEventRepository starred
MemberEventCollaborator added/removed
PublicEventRepository made public
ReleaseEventRelease published/created/deleted
WorkflowRunEventGitHub Actions run

Observations (from GitHub API, Local Git, Wayback, Vendors)

TypeDescriptionSources
CommitObservationCommit metadata and filesGitHub, Git, GH Archive
IssueObservationIssue or PRGitHub, GH Archive
FileObservationFile content at refGitHub
BranchObservationBranch HEADGitHub
TagObservationTag targetGitHub
ReleaseObservationRelease metadataGitHub
ForkObservationFork relationshipGitHub
SnapshotObservationWayback snapshotsWayback
IOCIndicator of CompromiseVendor
ArticleObservationSecurity report/blogVendor

IOC Types

from src import EvidenceSource, IOCType
from src.schema import IOC, VerificationInfo
from pydantic import HttpUrl
from datetime import datetime, timezone

# IOCs are created directly as schema objects
ioc = IOC(
    evidence_id="ioc-commit-sha-abc123",
    observed_when=datetime.now(timezone.utc),
    observed_by=EvidenceSource.SECURITY_VENDOR,
    observed_what="Malicious commit SHA found in vendor report",
    verification=VerificationInfo(
        source=EvidenceSource.SECURITY_VENDOR,
        url=HttpUrl("https://vendor.com/report")
    ),
    ioc_type=IOCType.COMMIT_SHA,
    value="678851bbe9776228f55e0460e66a6167ac2a1685",
)

Available IOC types: COMMIT_SHA, FILE_PATH, FILE_HASH, CODE_SNIPPET, EMAIL, USERNAME, REPOSITORY, TAG_NAME, BRANCH_NAME, WORKFLOW_NAME, IP_ADDRESS, DOMAIN, URL, API_KEY, SECRET

Testing

Run Unit Tests

cd .claude/skills/github-forensics/github-evidence-kit
pip install -r requirements.txt
pytest tests/ -v --ignore=tests/test_integration.py

Run Integration Tests (Optional)

Integration tests hit real external services (GitHub API, BigQuery, vendor URLs):

# All integration tests
pytest tests/test_integration.py -v -m integration

# Skip integration tests in CI
pytest tests/ -v -m "not integration"

Note: GitHub API integration tests use 60 req/hr unauthenticated rate limit. BigQuery tests require credentials (see below).

GCP BigQuery Credentials (for GH Archive)

GH Archive queries require Google Cloud BigQuery credentials. Two options:

Option 1: JSON File Path

export GOOGLE_APPLICATION_CREDENTIALS=/path/to/credentials.json

Option 2: JSON Content in Environment Variable

Useful for .env files or CI secrets:

export GOOGLE_APPLICATION_CREDENTIALS='{"type":"service_account","project_id":"...","private_key":"..."}'

The client auto-detects JSON content vs file path.

Setup Steps

  1. Create a Google Cloud Project
  2. Enable BigQuery API
  3. Create a Service Account with BigQuery User role
  4. Download JSON credentials
  5. Set GOOGLE_APPLICATION_CREDENTIALS env var

Free Tier: 1 TB/month of BigQuery queries included.

Requirements

pip install -r requirements.txt
  • pydantic - Schema validation
  • requests - HTTP client
  • google-cloud-bigquery - GH Archive queries (optional)
  • google-auth - GCP authentication (optional)

平台分布

Claude Code

26.18%
按下载量换算37

Antigravity

21.97%
按下载量换算31

Gemini CLI

17.56%
按下载量换算25

OpenCode

13.34%
按下载量换算19

trae

7.86%
按下载量换算11

windsurf

3.46%
按下载量换算5

安全审计

Gen Agent Trust Hub

可疑

Socket

通过

Snyk

可疑

权限和风险

需要联网

该 Skill 可能需要联网访问来源站点、仓库或外部 API;具体网络访问范围需要结合源码和 README 复核。

安装前确认

本站仅展示第三方公开信息,不托管安装包,不提供自动安装或运行环境。安装前应自行审查源码、依赖和命令行为。来源安全扫描存在提醒或未通过项,安装前请自行复核。

继续浏览同类 Skills