본문으로 건너뛰기
cccskills
무료GitHub에 공개

general-web-research

Conduct reproducible web searches and decision-oriented research across official sites, search engines, news, papers, forums, social platforms, and public datasets. Use when Codex must investigate a person, organization, product, market, technical resource, reputation, career choice, or other current topic; compare candidates; verify claims; collect dated citations; distinguish facts from inference and anecdote; or write an evidence-backed research report.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

설치 방법 보기

포함된 파일(3)

  • SKILL.md4.8 KB
  • agents/openai.yaml270 B
  • references/evidence-and-output.md2.2 KB

SKILL.md(원문)

설치하기 전에 에이전트에게 전달되는 지시 내용을 확인할 수 있습니다.

General Due Diligence

Turn an open-ended question into a traceable evidence base and a decision-ready due-diligence report.

Establish the Decision

  1. Restate the user's actual decision, constraints, deadline, and risk tolerance.
  2. Separate research questions into:
    • facts that can be verified;
    • estimates that require assumptions;
    • opinions or reputation claims;
    • unknowns that require direct contact.
  3. Define comparison criteria before collecting candidates. Weight criteria only when ranking helps the decision.
  4. Record the research date and treat fast-changing facts as time-sensitive.

For substantial work, read references/evidence-and-output.md before searching.

Build a Search Plan

Create a query matrix covering entities, aliases, languages, date ranges, and source types. Search broad-to-narrow:

  1. Discover vocabulary, candidates, and likely primary sources.
  2. Search official pages, institutional records, papers, grants, repositories, job postings, and public datasets.
  3. Search independent reporting and reputable aggregators.
  4. Search forums and social platforms for leads and lived experience.
  5. Run disconfirming queries such as complaints, departures, retractions, unavailable resources, and contradictory figures.

Use domain filters and quoted phrases. Search in every relevant language. Do not mistake search-result snippets for evidence; open the underlying page.

Collect Evidence

Maintain a source ledger while searching. For every material claim capture:

  • claim or data point;
  • source title, publisher, URL, publication/update date, and access date;
  • source tier and whether it is primary or secondary;
  • exact supporting passage or table location when practical;
  • caveats, conflicts, and freshness.

Prefer stable, direct URLs. Preserve enough context to audit the claim. Never include secrets, private personal data, or content obtained by bypassing access controls.

Evaluate Sources

Use this default hierarchy, adjusted for the question:

  1. Tier A: official records, first-party documentation, original papers/data, direct public statements.
  2. Tier B: reputable independent reporting, established databases, systematic surveys.
  3. Tier C: professional profiles, conference biographies, maintained aggregators.
  4. Tier D: forums, anonymous reviews, reposts, unverified social posts.

Assess authority, proximity, independence, date, specificity, and corroboration. A first-party source is strong for what an organization says or lists, but weak for independently judging quality.

For reputation or conduct claims:

  • report patterns, not isolated allegations;
  • corroborate across independent sources where possible;
  • distinguish direct experience from hearsay;
  • use neutral language and include uncertainty;
  • omit unnecessary personal details;
  • never convert silence into a positive finding.

Synthesize

Label statements as verified fact, reported claim, inference, or unknown when ambiguity matters.

Resolve conflicts explicitly. Prefer the more direct and current source, but preserve meaningful disagreement. Do not manufacture a single number from incompatible salary definitions, hardware counts, or date ranges. Show ranges and assumptions.

For rankings, publish the criteria and evidence behind each score. Treat missing data as unknown, not zero. Add a sensitivity note when small weighting changes alter the result.

Write the Deliverable

Lead with the answer and implications. Include:

  1. executive conclusion;
  2. scope, date, and method;
  3. evidence-backed findings;
  4. comparison table when useful;
  5. risks, contradictions, and unknowns;
  6. recommended next actions, including questions to ask people directly;
  7. source ledger or numbered references with live URLs.

Cite each material factual claim near the claim. Avoid a detached link dump. State what was not found and which access limits affected the work.

Quality Gate

Before finishing, verify:

  • every decisive claim has a citation;
  • current claims have recent sources or an explicit date caveat;
  • rankings do not hide missing evidence;
  • anecdotes are not presented as representative facts;
  • source links resolve to the claimed material;
  • the report answers the user's decision rather than merely summarizing pages;
  • actionable follow-ups address the largest remaining unknowns.

리뷰

아직 리뷰가 없습니다. 사용해 본 소감을 남겨 주세요.

댓글

로그인 없이 쓸 수 있습니다. 사용 후기, 질문, 작동하지 않은 환경 등을 편하게 남겨 주세요.

아직 댓글이 없습니다. 첫 댓글을 남겨 보세요.

보내면 바로 공개됩니다. 이메일 주소나 전화번호 같은 개인 정보는 쓰지 마세요. URL은 2개까지입니다.

0 / 1,000자

이 양식은 reCAPTCHA로 보호되며 Google 개인정보처리방침 및 서비스 약관이 적용됩니다.

같은 저장소의 스킬

Design original, scalable logos and production-ready asset sets for a specific usage context. Use when Codex needs to create or refine a logo, symbol, wordmark, app icon, repository identity, CLI mark, website brand, social preview, or Codex skill/plugin icon; adapt concepts, formats, dimensions, variants, and delivery checks to the target platform instead of producing a context-free image.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

ETOLucy/codex-workflow-skills102026년 7월 23일 업데이트

Discover, compare, and compose existing open-source GitHub repositories for a user's goal, using GitHub MCP for live discovery and verification plus Context7 semantic retrieval over repository code and official documentation. Use when a user asks for GitHub repo recommendations, open-source alternatives, a multi-repository solution stack, project comparison, or help choosing an implementation to adopt.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

ETOLucy/codex-workflow-skills102026년 7월 23일 업데이트

Teach software development through active project work while still producing a working result, maintaining bilingual English and Chinese project documentation, protecting private context, and publishing reviewed updates to GitHub only after confirmation. Use when Codex helps the user create, build, extend, debug, refactor, document, publish, or understand a software project and the user's goal includes learning, mentorship, guided implementation, pair programming, understanding the code, or becoming able to do similar work independently.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

ETOLucy/codex-workflow-skills102026년 7월 23일 업데이트

Audit, rename, and polish a GitHub repository across its local checkout and remote settings. Use when Codex needs to synchronize a local and GitHub repository name, set the description, homepage URL, or topics, create repository branding and a GitHub social preview, restructure a README, or prepare and publish a cohesive repository identity. Uses GitHub CLI for supported remote operations and delegates visual identity work to $create-context-aware-logos.

한국어 요약은 준비 중입니다. 원문 설명을 표시하고 있습니다.

ETOLucy/codex-workflow-skills102026년 7월 23일 업데이트

ETOLucy의 스킬 모두 보기

이 스킬의 문제 신고