Token导航 LogoToken导航TokenDH.com
研究检索操作浏览器github未标认证来源可访问许可证需确认审计提醒

product-implementproduct 实现

Agent Skill

product-implement 用于查找、检索和筛选相关信息,适合在 Codex、Claude、Cursor、Gemini CLI 中需要根据关键词、任务场景或来源线索快速定位候选结果时使用。可结合来源仓库、安装命令和原始 README 继续核验具体用法。安装前建议确认权限范围、维护状态,以及是否会触发联网、命令执行或文件读写。

总安装

196

周安装

8

GitHub Stars

公开资料未说明

下载量

63
CodexClaudeCursorGemini CLI

安装说明

本站只整理中文说明和来源信息,不托管安装包,也不代用户安装。

GitHub

来源数

2

许可证

unknown

最后核验

2026-05-01

来源状态

来源可访问

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:product-implement(product 实现)
来源仓库:https://github.com/jihunkim0/jk-skills
仓库路径:skills/product-implement
安装命令:
npx skills add https://github.com/jihunkim0/jk-skills --skill product-implement
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。该命令会通过 npx skills 从第三方来源获取 Skill;本站只展示命令,不托管安装包,也不自动执行。

skills.shnpx skills
npx skills add https://github.com/jihunkim0/jk-skills --skill product-implement

简介

用于查找、检索和筛选相关信息。适用宿主包括 Codex、Claude、Cursor、Gemini CLI,接入前应确认版本、权限和运行环境要求。

  • 适合根据关键词、任务场景快速定位候选结果。
  • 安装前建议确认权限范围和维护状态, 以及是否会触发联网或文件读写。
  • product-implement 属于研究检索类 Skill,可作为该场景下的辅助能力补充。

SKILL.md

Product Implement: The Factory Floor

You are the Controller Agent. You orchestrate ticket execution by dispatching worker subagents into isolated git worktrees. You do NOT write code — you coordinate.

Scripts: go run../../scripts/cmd/<name> (relative to this skill directory). References: Read on demand — only load when that phase is active.

Environment Detection

PiGemini CLI
Detectionsubagent tool availableNo subagent tools
Dispatchsubagent(cwd: "<worktree>")gemini --worktree + tmux/background
Resumesubagent_resume(sessionPath:...)gemini --resume in worktree

Phase 0: Warm Start

If .agents/products/.features/<slug>/context-bundle.md exists, read it for instant context. Otherwise, generate one using references/context-bundle.md protocol.

Run go run../../scripts/cmd/assess state --slug=<slug> → AssessResult JSON.

StatusAction
RESUMERead references/resume-protocol.md, follow steps
COMPLETEInform user. Offer feature-complete ceremony → references/feature-complete.md
FRESHProceed to Phase 1

Also check git worktree list — active feat/* worktrees = in-flight work. Offer to resume.

Phase 0.5: Events — Implementation Started

Emit: go run../../scripts/cmd/events emit --slug=$SLUG --skill=implement --type=phase_start --payload='{"phase":"implementation","issue_id":"<id>"}'

Phase 1: Issue Selection & Sprint Contract

  1. Detect tracker: Linear → linear api '{issues(...)}' (GraphQL — authoritative). GitHub → gh issue list --state all
  2. Present ALL issues to user (ID, title, state). Cross-reference with .agents/products/plans/<slug>.md
  3. No issues? Plan exists → "Run /product-issues". No plan → "Run /product-plan". Nothing → "Run /product-pipeline"
  4. Lock: Linear --state started. GitHub --add-assignee @me
  5. Sprint contract: go run../../scripts/cmd/contract create --slug=<slug> --issue-id=<id> --title="<title>" --criteria='["..."]' --scenarios='["..."]'
  6. Learnings: go run../../scripts/cmd/compound retrieve --slug=<slug>
  7. Journal: go run../../scripts/cmd/journal append --slug=<slug> --skill=implement --phase=implementation_started --action=complete

Commit Discipline

Make granular, atomic commits throughout implementation:

  • Each passing test: git commit -m "test: <what it tests>"
  • Each feature increment: git commit -m "feat: <what it adds>"
  • Each refactor: git commit -m "refactor: <what changed>"
  • Each gate fix: git commit -m "fix: <gate finding addressed>"
  • WIP commits every 5 minutes during long implementations

Phase 2: Dispatch Implementers

  1. git worktree add.worktrees/feat-<ID>-<desc> -b feat/<ID>-<desc>
  2. Build prompt from assets/implementer-template.md — substitute contract, issue body, learnings, prior attempts
  3. If .stitch/DESIGN.md exists → inject into {{DESIGN_MD}} slot (implementer must use design tokens for all UI)
  4. go run../../scripts/cmd/registry spawn --agent-id=<id> --issue-id=<id> --worktree=<path>
  5. Dispatch: Pi → subagent(name: "Impl: <ID>", cwd: "<worktree>"). Gemini → tmux/background
  6. For parallel-safe tickets, dispatch multiple simultaneously

Phase 3: Review Gates

Read references/gate-protocol.md — 9-gate system with parallelization and blocking rules.

Execution

  1. Gate 1 (Tests): Dispatch per references/gate-tests.md — run full test suite → PASS/FAIL. Blocks everything
  2. Gates 2–3, 5–9: Dispatch review sub-agents per gate protocol (Gate 4 runs in Phase 4). Each gate has its own reference file:

- Gate 2: references/gate-spec.md · Gate 3: references/gate-quality.md - Gate 5: references/gate-time-complexity.md · Gate 6: references/gate-space-complexity.md - Gate 7: references/gate-logic-optimality.md · Gate 8: references/gate-test-strategy.md - Gate 9: references/gate-design-quality.md (only if ticket produces UI)

  1. Persist results → .agents/products/.gates/<slug>-<issue>-<gate>.json (schema: assets/gate-findings-schema.json)
  2. Comment on tracker: go run../../scripts/cmd/tracker write-comment --id=<issue-id> --body="<comment text>"

On Failure

  1. go run../../scripts/cmd/wtf-likelihood --slug=<slug> --issue-id=<id>CONTINUE | ESCALATE | ABORT
  2. CONTINUE → consolidate .gates/ findings, re-dispatch implementer with feedback
  3. ESCALATE → present to user with WTF score
  4. ABORT → hard stop, report findings to user
  5. Hard cap: 5 round trips regardless of WTF score

Phase 4: Polish

  1. Gate 4 (Cleanup): Dispatch per references/gate-cleanup.md — subtraction testing, dead code removal
  2. optimize-code: Scoped to changed files, deletion-first
  3. document-update: Sync docs with changes
  4. Gate 9 visual loop (if UI + browser available): max 10 iterations screenshot → fix → re-check
  5. Journal: go run../../scripts/cmd/journal append --slug=<slug> --skill=implement --phase=polish_complete --action=complete

Phase 5: Ship

Read references/ship-checklist.md — verify every item before shipping.

  1. go run../../scripts/cmd/contract validate --slug=<slug> --issue-id=<id> — all criteria met
  2. go run../../scripts/cmd/features update --slug=<slug> --id=<feature-id> --status=complete
  3. go run../../scripts/cmd/compound capture --slug=<slug> --title="Issue <id>: learnings" --domain=implementation --body="<WIP learnings>" — promote WIP learnings
  4. go run../../scripts/cmd/registry complete --agent-id=<agent-id> --result=success
  5. User ships: PR (body includes Fixes <Issue-ID>), merge, worktree cleanup
  6. Journal: go run../../scripts/cmd/journal append --slug=<slug> --skill=implement --phase=shipped --action=complete
  7. Emit: go run../../scripts/cmd/events emit --slug=$SLUG --skill=implement --type=phase_end --payload='{"issue_id":"<id>"}'

More tickets? → Loop to Phase 1. All done? → Read references/feature-complete.md, run the 7-step ceremony.

Note: handoff write/read available for structured data transfer between skills. Currently handled via journal resume-hints. Use handoff for complex data payloads.

适合场景

01

用户想查找某类 Agent Skill 时

02

需要根据任务场景推荐可安装能力包时

03

需要对比不同来源的安装命令和来源信息时

能力概览

能力 1

按任务关键词查找相关 Skills

能力 2

展示可复制的安装命令

能力 3

保留来源站点、仓库和原始说明,方便继续核验

能力 4

展示第三方安全扫描或审计结果

安装后应在对应宿主中按原始 README 的触发条件使用;具体调用方式请以来源页面和 README 为准。

平台分布

Codex

34.77%
按下载量换算22

Claude

29.37%
按下载量换算19

Cursor

16.89%
按下载量换算11

Gemini CLI

9.58%
按下载量换算6

安全审计

Gen Agent Trust Hub

通过

Socket

通过

Snyk

可疑

权限和风险

操作浏览器

该 Skill 可能涉及浏览器控制能力,使用时可能读取或操作网页内容,需要在受控环境中确认权限边界。

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 warning/failed 结果,不能写成本站确认安全。当前只有一个来源,正式发布前建议补源仓库或其他目录站核验。

来源信息

继续浏览同类 Skills