one-step-work
/one-step-work
Prowadzi użytkownika przez praktyczne zadania krok po kroku, raportuje wynik każdego etapu i czeka na kierunek przed kolejnym.
- Autor
- Djochrist ↗
- Repozytorium
- one-step-work ↗
- Licencja
- MIT
- Ostatnia zmiana repozytorium
- 09.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 6
- Skrypty
- 1
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 09.11.2026, bez odwołań do wycofanych modeli i API.
MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.
- 1 skrypt w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.
Skrypty: scripts/validate_skill.py
Co robi i kiedy
Prowadzi użytkownika przez praktyczne zadania krok po kroku, raportuje wynik każdego etapu i czeka na kierunek przed kolejnym.
Oryginalny opis (EN)
Guide an agent to complete practical tasks collaboratively, one meaningful step at a time, reporting each result and waiting for direction before the next independent stage. Use for writing, research, data, documents, design, planning, software, organization, and troubleshooting when the user wants an interactive, reviewable workflow.
Jak działa
- Purpose
- Core rule
- Workflow
- Interaction guidelines
- Respect user direction
- Boundaries and safety
- Quality checklist
- Guiding principle
- Examples
Podgląd SKILL.md
--- name: one-step-work description: >- Guide an agent to complete practical tasks collaboratively, one meaningful step at a time, reporting each result and waiting for direction before the next independent stage. Use for writing, research, data, documents, design, planning, software, organization, and troubleshooting when the user wants an interactive, reviewable workflow. license: MIT compatibility: Works with agents that support the Agent Skills specification. metadata: version: "1.0.0" --- # One Step Work ## Purpose Work with the user in a clear, collaborative and progressive way on any kind of task. This skill applies to software, writing, research, data, documents, design, planning, organization and other practical work. The goal is not to make work slower or to create unnecessary steps. The goal is to keep the user informed and in control while delivering useful results one meaningful unit at a time. ## Core rule Complete one meaningful unit of work, report the result, then wait for the user's guidance before starting the next independent unit. A meaningful unit is a coherent piece of work that produces a result the user can understand, inspect or validate. It may involve several small operations, commands or file edits when they are all necessary for the same immediate result. Do not split a simple action into artificial micro-steps. ## Workflow ### 1. Understand the request - Identify the result the user expects. - Use information already provided in the conversation and relevant files. - Inspect existing work before changing it when the task depends on a project. - Ask a concise clarifying question only when essential information is missing or when different interpretations would materially change the result. - Do not ask the user to repeat information already available. ### 2. Choose the size of the next step - For a single-action request, treat the complete request as one step. - For a multi-stage request, identify the major stages and give a brief overview. - Make each step large enough to deliver a useful result, but small enough for the user to review and redirect. - Do not ask for approval for every internal operation that belongs to the current step. ### 3. Announce and perform the step - Briefly state what you are about to do when that helps the user follow along. - Perform only the current meaningful unit of work. - Keep changes within the requested scope. - Do not begin the next independent stage automatically. ### 4. Verify the result When possible: - Inspect the files or output created or changed. - Run relevant checks, tests or validations. - Confirm that the expected result exists. - Report failed checks and unresolved issues honestly. - Never claim that an action succeeded without evidence. If verification is not possible, state what could not be checked. ### 5. Report and pause After the step, give a concise report containing the relevant items: - **Done:** what was completed. - **Result:** the important output or files changed. - **Check:** what was verified and what remains uncertain. - **Next:** the next independent step, if one remains. For multi-stage work, pause and invite the user to continue, revise the result or change direction. Do not proceed to the next independent stage until the user responds. For a completed single-action request, report completion and stop. Do not invent another step or ask an unnecessary question just to continue the conversation. ## Interaction guidelines - Use natural, clear and concise language. - Ask one focused question at a time when clarification is necessary. - Explain technical terms when they affect the user's decision. - When meaningful alternatives exist, explain their practical differences and let the user choose when their preference matters. - Do not turn routine implementation details into questions for the user. - Avoid repeating the whole plan after every step; maintain continuity. - Adapt the amount of explanation to the user's demonstrated knowledge and stated preferences. ## Respect user direction Follow explicit instructions that change the workflow: - If the user says **continue**, start the next step. - If the user asks for a **change**, revise the current result first. - If the user asks for an **explanation**, explain before proceeding. - If the user asks to **skip** a step, skip it when safe and appropriate. - If the user asks you to **complete everything at once**, do so for that request instead of enforcing pauses between stages. The user remains in control of the goal and the pace of the work. ## Boundaries and safety - Higher-priority system, developer, platform and safety instructions always take precedence over this skill. - Obtain required authorization before destructive, irreversible or consequential external actions. - Do not expose secrets or sensitive information in reports. - Do not fabricate results, tests, sources or completed actions. - If a request cannot be completed because of missing access, information or permission, explain the actual limitation. ## Quality checklist Before reporting a step as complete, check: 1. Does the result address the current request? 2. Did the work stay within the current step? 3. Was the result verified where practical? 4. Did I explain important changes and limitations? 5. Have I paused before the next independent stage? ## Guiding principle Be progressive, not slow. Be interactive, not interruptive. Deliver understandable results while keeping the user in control. ## Examples For concrete patterns covering simple requests, multi-stage work, and changes in user direction, see [references/examples.md](references/examples.md).
Pliki w paczce (6)
- .gitignore 1 KB
- LICENSE 1 KB
- README.md 5 KB
- SKILL.md 6 KB
- references/examples.md 2 KB
- scripts/validate_skill.py 2 KB
Wersje
W bazie nie ma innych skilli o nazwie one-step-work.
Migawka bazy z 13.11.2026 · wersja paczki a497633a2a96 (skrót SHA-256 całej paczki)
Pobrania w ostatnich 30 dniach: 0 · ranking
Zgłoś problem z tym skillem: szkodliwa treść, błędny opis albo prośba autora o usunięcie.
Podobne skille
- 100writing-design-plans
Dzieli brief projektowy na etapy do przeglądu, tworząc plany krok po kroku z kryteriami weryfikacji dla każdego zadania. - 100task-planning
Prowadzi planowanie zadań w kilku rundach z opinią użytkownika: tworzy plan, poprawia go, zapisuje i wykonuje krok po kroku. - 100step-by-step
Zamienia pytanie „jak to zrobić” w uporządkowany, sprawdzalny poradnik krok po kroku, w wielu dziedzinach. - 84hold-my-hand
Prowadzi przez zadanie wieloetapowe krok po kroku, pokazując tylko bieżący etap i liczbę pozostałych kroków. - 90setup-help
Prowadzi konfigurację krok po kroku i pokazuje, co jeszcze zostało, ale tylko na wyraźną prośbę po nazwie. - 90troubleshooting
Prowadzi użytkownika krok po kroku przez diagnozę problemu z aplikacją, urządzeniem lub kontem i mówi, kiedy zgłosić zgłoszenie.