Works with the AI Agents you already use

    Claude CodeClaude CodeCursorCursorCodex CLICodex CLIDeepSeekDeepSeekMuseMuse+22 more

    Solo Dev Kit: Session Summary, Push Gate, /record

    1

    For one person building a product with Claude Code: a session-start hook that computes project state from git and three ledgers every time, a /record skill that logs mistakes into the file that owns them, and a gate that runs typecheck and lint before git push.

    Free

    Security scanned
    Solo Dev Kit: Session Summary, Push Gate, /record

    Solo Dev Kit: Session Summary, Push Gate, /record

    Example session with this skill installed

    이 프로젝트에 1인 개발 틀 깔아줘. (Set up the solo dev kit in this project.)

    • Read your context and instructions
    • Compiled the solo dev kit

    Summary

    Looked at the project first (no CLAUDE.md, no ledgers, package.json has typecheck and lint), showed the plan, installed after your yes, and ran both hooks once. Nothing existing was overwritten.

    Plan (install.mjs, real output)

    프로젝트: my-app  (계획만 — 깔려면 --apply)
    
    파일:
      새로  .claude/hooks/pre-push.mjs
      새로  .claude/hooks/session-start.mjs
      새로  .claude/rules/example.md
      새로  .claude/skills/record/SKILL.md
      새로  CLAUDE.md
      새로  docs/changelog.md
      새로  docs/decisions.md
      새로  docs/milestones.md
    
    .claude/settings.json 훅:
      보탬  SessionStart [startup|resume|clear|compact]
      보탬  PreToolUse [Bash] if Bash(git push *)
      보탬  PreToolUse [Bash] if Bash(git -C * push *)
    
    푸시 전 검사(순서대로, 첫 실패에서 멈춘다):
      1. npm run typecheck
      2. npm run lint
    

    (새로 = new, 보탬 = add, 푸시 전 검사 = checks before push.)

    You: 좋아, 깔아

    Session summary after a few days of use (real output from a test project)

    [자동 요약 — 편집하지 말 것. 상태를 적는 곳은 docs/milestones.md · docs/decisions.md · docs/changelog.md 뿐]
    git: main · 변경 1개 · 8cc9383 2026-11-01 feat: a · 기록 뒤 커밋 1개
    마일스톤: 끝 1/2 · 남은 것 M2(◐)
    사람 손: 없음
    결정: 열린 것 1건(3) — 임의로 정하지 않는다 · 정한 것 2건 · 다음 번호 4
    변경 기록(1건): 10-28 결제를 뒤로 미뤘다
    절차: 한 덩어리가 끝나면 /record <제목> — 실수는 그 스킬 · 규칙 파일의 '지난 실수'에 적는다(/record가 묻는다)
    

    (Milestones 1 of 2 done, M2 in progress; 1 open decision #3, next number 4; latest change: payment postponed; 1 commit since the last record.)

    Push gate

    git push with a failing lint -> blocked with exit 2 and the first lines of the lint output. git commit -m "push it" and git stash push pass through untouched.

    Next steps

    • Fill the <...> blanks in CLAUDE.md together (what to read first, what not to do).
    • Move any TODO or decision notes you already had into the ledgers.
    • Commit when you are ready, e.g. chore: solo dev kit (session summary, push gate, /record).
    Looked at the project first (no CLAUDE.md

    Connects securely to your tools. The creator never sees your data.

    About this skill

    Working alone with an agent, the same three things go wrong every week: a status line in CLAUDE.md that was true a month ago, a mistake the agent already made once and makes again, and a push that breaks the build because nothing ran lint. This kit puts a file in each of those places.

    The three ideas

    1. State is computed, not written. A SessionStart hook reads git (branch, changes, last commit, commits since the last record), docs/milestones.md (done / doing / waiting), docs/decisions.md (open decisions and the next number) and the newest row of docs/changelog.md, and prints a 6-line summary at the top of every session. Nothing to keep in sync.
    2. Mistakes live in the file that owns the work. /record <title> first asks "was there a mistake this round?" and writes it into that skill's or rule's ## Past mistakes section, so the next session that opens the file reads it. Then it updates the ledgers and makes a docs commit.
    3. Push gate. A PreToolUse hook catches git push and runs preflight / typecheck / lint from package.json (whichever exist), stopping at the first failure with the reason. It picks npm / pnpm / yarn / bun from the packageManager field or the lockfile.

    Plus: open decisions get numbers and the agent does not decide them on its own; path-scoped rules (paths: frontmatter) load only when matching files are opened.

    Install flow

    install.mjs <project> prints a plan only: new files, files that already exist (same / different, with a diff -u command), sections missing from your CLAUDE.md, hooks to add to .claude/settings.json, and the commands the gate will run (or "no checks" if there are none). With --apply it copies without overwriting and merges hooks into settings.json without removing yours; running it twice changes nothing.

    Zero dependencies (Node 18+, git). Templates and instructions are in Korean; the hooks work on any project.

    Source: https://github.com/Stophyuk/solutionzip-skills

    Find more skills like this on the Claude Skills Marketplace, the Codex Skills Marketplace and the Gemini Skills Marketplace, or browse Claude Skills for Agents & Orchestration.

    The hooks are Claude Code specific (SessionStart, PreToolUse); other agents get the ledgers and /record only. The push gate runs only fast checks (under about 280 seconds total); builds and slow tests belong in CI. The settings 'if' filter needs a recent Claude Code; the script also re-checks the command. Not tested on Windows. Templates and instructions are in Korean.

    How to use

    1. 1

      Connect your agent to the Agensi MCP

      Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup

    2. 2

      Sign in to Agensi

      Your agent shows a sign-in link. Sign in and click Allow.

    3. 3

      Ask for what you need

      Your agent loads this skill and uses it on your task. It always asks before buying a skill.

    Reviews

    No reviews yet

    Be one of the first to try it. Every listed skill goes through the trust checks below.

    Security scanned

    Checked by our 8-point scan before listing

    Fresh listing

    Recently published to Agensi

    Free to use

    No account required to browse

    Trust & safety

    Security scan

    Security scan passed 16 days ago

    • Free with an Agensi account

    Software
    Claude CodeNode.js 18+Git

    Permissions

    Terminal / Shell
    Write Files
    Read Files
    Listed16 days ago

    What's inside

    Frequently Asked Questions

    No. The installer prints a plan first, never overwrites an existing file (it shows a diff command instead), and merges hooks into settings.json while keeping your hooks and permissions. If settings.json is not valid JSON it leaves it alone.

    The plan says 'no checks' and the agent asks which checks to add, either as package.json scripts or in the EXTRA list inside pre-push.mjs.

    Yes. Point the CONFIG object at the top of session-start.mjs to your files and section names, or move the items into the ledgers. Status cells only need to start with one of four marks (done, doing, waiting, not started).

    It is a Claude Code hook, so git push --no-verify does not bypass it. The skill tells the agent to fix the failure and push again, not to disable the check.

    New in Agents & Orchestration

    Free