Przejdź do treści
Biblioteka skilli
Zarządzanie i produkt · Planowanie

plan

/plan

Układa plan większej implementacji, odpowiada na jawne prośby o planowanie i raportuje status autoryzowanej pracy.

Repozytorium
skills ↗
Licencja
MIT
Ostatnia zmiana repozytorium
12.11.2026

Ocena i bezpieczeństwo

100/ 100Klasa: dobry
Format25 / 25
Frontmatter z nazwą i opisem, poprawna nazwa, opis do 1024 znaków, nazwa zgodna z folderem.
Odkrywalność25 / 25
Opis mówi, kiedy użyć skilla, ma sensowną długość i jest w 3. osobie.
Treść35 / 35
Co najmniej 150 słów i do 500 linii, nagłówki, listy kroków, przykłady albo kod.
Higiena15 / 15
Bez zaślepek typu TODO, bez ścieżek Windows, z licencją.
Ryzykobrak sygnałów w skanie
Pliki w paczce
2
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 12.11.2026, bez odwołań do wycofanych modeli i API.

Licencja

MIT, otwarta: można używać i zmieniać, z zachowaniem informacji o autorze.

Co robi i kiedy

Układa plan większej implementacji, odpowiada na jawne prośby o planowanie i raportuje status autoryzowanej pracy.

Oryginalny opis (EN)

Use to plan substantial implementation, answer an explicit planning request, or report or record status against authorized work.

Jak działa

  1. 1. Establish the outcome and evidence
  2. 2. Keep one plan folder per task
  3. 3. Reach agreement and propose the plan
  4. Track authorized work
Sekcje z nagłówków SKILL.md, w oryginale · 46 linii, 1 190 słów
Podgląd SKILL.md
---
name: plan
description: "Use to plan substantial implementation, answer an explicit planning request, or report or record status against authorized work."
---

# Plan

Produce an evidence-based plan for the outcome the human wants. Product decisions and business rules belong to the human. Use this skill when requested or when implementation involves consequential risk, unresolved decisions, or multiple dependent stages. File count alone does not make work substantial. A read-only review or investigation does not start a Plan record just because it is substantial. See the [routing guide](../../references/Routing.md) when ownership is unclear.

For a status request on authorized work, go directly to [Track authorized work](#track-authorized-work).

## 1. Establish the outcome and evidence

- Identify the desired outcome, scope, constraints, stakeholders and observable measures of success. Separate decisions from proposals and open questions; do not invent requirements or business logic. Preserve established behavior outside the requested change using current evidence. Ask the human when a business rule is affected, genuinely uncertain, or requires a product decision. For example, an unchanged, evidenced delivery fee needs no separate confirmation when only the free-shipping threshold changes; if the fee itself is affected or unclear, surface the decision.
- Inspect the relevant current implementation, connected behavior, tests, documentation and project conventions before proposing changes. Reuse established behavior and mechanisms where they fit. Cite the relevant paths or other evidence in the planning files; identify missing or inaccessible evidence.
- Compare the human's factual statements with the evidence. When they conflict, explain the discrepancy respectfully and ask how to proceed where it changes the decision. A deliberate hypothetical, simulation or explicit product decision may be planned as such, but must not be recorded as a verified fact about the current implementation.
- Ask focused questions for unresolved product choices, business rules, success criteria or material scope. Offer evidence-based alternatives and a recommendation where useful. Continue independent investigation while an answer is pending; do not decide a business rule for the human or treat silence as agreement. Choose routine verification checks that follow from established behavior or agreed criteria. Use [Test](../test/SKILL.md) when selecting software checks needs its guidance.
- Use [Focus](../focus/SKILL.md) for an unresolved product choice or [Architect](../architect/SKILL.md) for a consequential technical design. Keep their evidence with the decision; sequence the chosen work here.
- When the outcome includes shipping or production changes, name the relevant owners and exit evidence: [Secure](../secure/SKILL.md) for security/privacy, [Migrate](../migrate/SKILL.md) for state transitions, [Release](../release/SKILL.md) for the candidate, [Deploy](../deploy/SKILL.md) for environment activation and [Operate](../operate/SKILL.md) for health/recovery. Include dependency order, irreversible boundaries and a usable recovery path where relevant. Do not turn these into mandatory stages for a local edit.

## 2. Keep one plan folder per task

At the project root, create a [numbered plan folder](../../references/Workspace.md#numbered-records) such as `.work/plans/0001-Sales-Contracts/`, or resume the existing folder for the same task. Preserve unrelated plans. Keep task-specific evidence alongside the plan. For substantial work, use these three files. For a small explicit plan request, you may use one `Plan.md` with `## Outcome`, `## Agreement`, and `## Plan` sections containing the same information.

- **`Outcome.md`** — desired outcome, scope and constraints; observable acceptance criteria and how the eventual work will be checked; unresolved criteria marked pending. Describe expected behavior without asserting that unfinished work is complete.
- **`Agreement.md`** — decisions and the alternatives considered, including the human's choice, reason and relevant evidence when known. Attribute decisions accurately; distinguish proposed, pending, agreed and superseded choices. Record unanswered questions and retain prior decisions and alternatives when they change. Do not invent alternatives or agreement; say when no alternative has yet been identified.
- **`Plan.md`** — an explicitly marked draft or approved detailed outline, grounded in the inspected implementation and conventions. For each meaningful step record the intended action, affected area, dependencies and verification against `Outcome.md`. Identify unresolved decisions and limits instead of presenting guesses as settled requirements.

Create the record while investigating, update it as evidence and decisions emerge, and preserve it for the human's historical reference. Resume existing records in their current format; no migration is needed. If the repository root or write access is unavailable, explain the blocker and ask for a suitable repository or access; do not silently store the plan elsewhere or claim the record exists.

## 3. Reach agreement and propose the plan

Before presenting the plan, check the Outcome, Agreement, and Plan for proposed business rules, changed success criteria, or material scope decisions presented as settled without human confirmation. Mark unresolved decisions pending and ask the human to decide. Record routine checks that verify established behavior as the agent's verification choices.

For a substantial plan, request a fresh read-only [Audit](../audit/SKILL.md) of the identified draft when the host permits delegation and it is authorized. Give the reviewer the goal, files, constraints and evidence; use [Orchestrate](../orchestrate/SKILL.md) when coordination is needed. Handle material findings with the [review loop](../../references/ReviewLoop.md), and record the reviewer, findings, repairs and focused recheck in `Agreement.md`. If independent review is unavailable, record and disclose the gap; label any self-check accurately. Follow any stronger review gate required by the human or host.

Present a concise brief of the intended actions, outcome and checks, relevant evidence or constraints, and unresolved product decisions. Link the plan record in its current format. Ask the human to resolve material choices; continue independent authorized work while an answer is pending. If the request is for a plan or approval is otherwise required, obtain approval before dependent execution.

Honor clear existing authorization to execute without asking for it again. Record what was authorized and by whom in the plan's Agreement record: `Agreement.md` or the `## Agreement` section of a compact `Plan.md`. Distinguish that authorization from explicit approval of a particular written plan. Mark `Plan.md` approved only when the human actually approves it; silence, elapsed time and the agent's recommendation are not approval. Before dependent work, resolve decisions that require the human and satisfy any stronger host or user gate. During execution, record routine technical adjustments within scope; return to the human for changes to the outcome, product decisions or authorized scope. Neither a plan nor its approval grants authority beyond the human's instructions and host permissions.

**Done:** The plan record describes the outcome, decisions, evidence and steps, with truthful authorization and review status. Report unresolved decisions or verification gaps; leave a written plan in draft until explicitly approved.

## Track authorized work

1. Identify the matching authorized work and plan record. Read `Outcome.md`, `Agreement.md` and `Plan.md` for a three-file plan, or the Outcome, Agreement and Plan sections of a compact `Plan.md`. Recover commitments and acceptance criteria. Check actual artifacts, checks or owner reports; distinguish direct, reported, stale and unavailable evidence. If no matching record establishes authorized work, report what you can verify and leave plan files untouched.
2. Give an as-of snapshot of completed and remaining work, milestone variance, blockers or dependencies, known owners and dates, and decisions needed. Link material evidence and mark unknowns. Surface proposed changes without treating them as approved commitments.
3. Return status questions read-only. When asked to save status, update only the top-level `## Current status` section in the matching `Plan.md` in either format, adding it if absent. Keep a single plan record and preserve its approval state. Assignments, external updates, and ongoing monitoring require their own authorization.

**Done:** Return the evidence-backed snapshot and link its location if saved.
Pliki w paczce (2)
  • SKILL.md 9 KB
  • agents/openai.yaml 1 KB

Wersje

RepozytoriumWynikLicencja
Ta strona · nazmirket/skills100MIT
mlophez/polix-ai100Apache-2.0
jeremylongshore/tons-of-skills-marketplace100MIT
duronext/jig100MIT
rjmurillo/ai-agents100MIT
rjmurillo/ai-agents100MIT
rjmurillo/ai-agents100MIT
xoai/sage100MIT
JoDi-27/JoDi-SDD100MIT

Migawka bazy z 13.11.2026 · wersja paczki 6a2c96f5c024 (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

  • 100plan-feature
    Prowadzi planowanie funkcji przez wymagania, badanie kodu, ocenę wykonalności i plan implementacji.
  • 90create-plan
    Przeprowadza wywiad z użytkownikiem i pisze plan wdrożenia dla bieżącej prośby o funkcję.
  • 100planning
    Prowadzi planowanie implementacji z zapisem w plikach task_plan.md, findings.md i progress.md.
  • 100issue-updater
    Aktualizuje zgłoszenia GitHub o planowanie, status i postęp implementacji w ramach aktywnego, jawnie wywołanego przepływu pracy.
  • 90planning-changes
    Przygotowuje zwięzły plan większej zmiany z kryteriami akceptacji, przypisaniem plików do osób, zależnościami i oczekiwanymi wynikami.
  • 100plan-steps
    Układa plan implementacji w małe zadania: test, kod, polecenie sprawdzające i status w plan.md, na podstawie prośby o kroki.

Uwagi i zgłoszenia

Wiadomość trafi do Bartosza Sokolińskiego. Adres e-mail jest potrzebny tylko do odpowiedzi. Szczegóły w polityce prywatności.