소스 정보
- 저장소
- gautam-achieveai/ClaudePlugins
- 최근 소스 활동
- 2026년 7월 9일 07:23
- 감지된 SKILL.md 언어
- 영어
- 스타
- 3
- 포크
- 0
설치 방법
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
소스 파일 검토
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
메뉴
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/gautam-achieveai/ClaudePlugins --skill draft-feature명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
Internal helper. Load only when explicitly named by another skill or agent. Publishes goal-aligned, deduplicated PR feedback with clear blocker outcomes and stable closure criteria so reviews converge without repeated comment rounds.
Conduct goal-aligned code reviews of individual pull requests, analyzing correctness, solution fit, performance, code alignment, testing coverage, and code quality while helping authors converge quickly on a mergeable change. Provides prioritized, actionable feedback with stable closure criteria. Use when asked to "review PR #[number]", "code review pull request", "check PR for issues", or "analyze PR changes". Works on GitHub or Azure DevOps, with PR numbers, branch names, or GitHub/Azure DevOps PR URLs. NOT for developer performance reviews over time.
This skill should be used when the user asks to "babysit a PR", "babysit my pull request", "monitor my PR", "watch my pull request", "keep my PR green", "fix PR build failures automatically", "handle PR review comments", or wants autonomous Azure DevOps PR monitoring that fixes build breaks, test failures, code coverage gaps, and review comments on a polling loop.
SOC 직업 분류 기준
SKILL.md 표시 중
| name | draft-feature |
| description | Internal helper. Load only when explicitly named by another skill or agent. |
| user-invocable | true |
| disable-model-invocation | false |
You derive high-quality feature requirements from a rough idea. You do NOT
create the work item and you do NOT detect the provider — the
development:draft-work-item router already resolved the provider and will
handle duplicate-check, preview, and creation. Your output is a composed
{type, title, body, meta} handed back to the router.
Inputs from the router: provider (GitHub or Azure DevOps), rawRequirement
(the user's original text), and priorAnswers (anything already volunteered —
never re-ask these).
Work the phases below in order. Phases loop: new understanding sends you back to re-ground earlier phases. Ask questions one at a time, multiple-choice where possible.
Build real grounding before asking anything:
Explore subagents over the codebase to find the components,
patterns, and existing conventions this feature would touch.searchWorkItems
for related/duplicate efforts already tracked.WebSearch only when the requirement leans on external/unfamiliar
tech, standards, or APIs.Summarize what you learned and where the requirement meets reality.
Turn the gaps between the high-level ask and the discovered context into
questions, asked one at a time. Skip anything in priorAnswers. When the
user's answers materially change your understanding, loop back to Phase 1 to
re-ground against the code. Stop when the core intent, users, and value are clear.
When ambiguities are resolved, dispatch the development:blind-spot-detector
agent with the feature lens, passing the working requirements plus the
context you gathered. It surfaces what your feature focus misses: edge cases,
cross-cutting impact, non-functional needs (perf, security/authz, privacy/EUII,
a11y, i18n, observability), data/migration/compat, and acceptance gaps.
You may dispatch multiple detectors in parallel for breadth on a larger feature.
Resolve what the scan surfaced. Fold confirmed gaps into the requirements; for open questions, ask the user (one at a time). If a finding reopens the core requirement, loop back to the relevant earlier phase (down to Phase 1).
Compose the body, applying the active provider's mention conventions (the router
supplies these; if working standalone, GitHub →
../draft-work-item/reference/gh-mention-conventions.md, ADO →
../draft-work-item/reference/ado-mention-conventions.md).
## Summary
<what to build>
## Value
<who benefits and why — the user/persona and the value delivered>
## Acceptance Criteria
- [ ] <verifiable, pass/fail>
- [ ] ...
Generate a concise title (< 80 chars). Every acceptance criterion must have a clear pass/fail signal.
Dispatch the code-reviewer:over-engineering-review agent against the drafted
requirements, focused on gold-plating, scope creep, and unrequested scope. When
the feature looks like it may overlap existing functionality, also dispatch
code-reviewer:duplicate-code-detector to catch duplicate effort.
Trim scope and fold in the review findings. Ask the user about any genuine ambiguity the review raised (one at a time). If the changes are substantial, loop back to the relevant phase to re-validate.
Return {type: feature/user-story, title, body, meta} to
development:draft-work-item. The router runs the duplicate check, shows the
mandatory preview, creates the item on the resolved provider, and offers the
follow-up. You never create the item yourself.