Przejdź do treści
Biblioteka skilli
AI i agenci · Planowanie

workflows-driven

/workflows-driven

Rozkłada pracę na subagentów lub fazy, gdy zadanie, takie jak audyt, migracja lub przegląd badawczy, przekracza pojemność jednego kontekstu, z wyłączeniem zmian zdalnych, publikacji i nieodwracalnych.

Licencja
Apache-2.0
Ostatnia zmiana repozytorium
28.09.2026

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
5
Skrypty
0
Komentarze HTML
0
Adresy URL w SKILL.md
0
Przestarzałość

Brak sygnałów. Repozytorium zmienione 28.09.2026, bez odwołań do wycofanych modeli i API.

Licencja

Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.

Co robi i kiedy

Rozkłada pracę na subagentów lub fazy, gdy zadanie, takie jak audyt, migracja lub przegląd badawczy, przekracza pojemność jednego kontekstu, z wyłączeniem zmian zdalnych, publikacji i nieodwracalnych.

Oryginalny opis (EN)

Use when work decomposes across subagents or phases: audits, migrations, research sweeps, or scale one context cannot hold. Not for remote, credential, publish, deploy, or irreversible changes.

Jak działa

  1. Contract
  2. Inputs
  3. Procedure
  4. Closing chain: Work -> Review-fix-loop -> Offensive Aggressive Simplify
  5. Failure and recovery
  6. Output
Sekcje z nagłówków SKILL.md, w oryginale · 55 linii, 942 słów
Podgląd SKILL.md
---
name: workflows-driven
description: 'Use when work decomposes across subagents or phases: audits, migrations, research sweeps, or scale one context cannot hold. Not for remote, credential, publish, deploy, or irreversible changes.'
---

# Workflows-driven

## Contract

| Field | Bound contract |
|---|---|
| Trigger | The work is an audit, a migration, a broad research or review sweep, or scale that one context cannot hold. |
| Authority | Reversible local: writes only named local evidence artifacts; rollback is a no-op because all mutable state is scoped to per-task evidence files owned exclusively by one worker. No remote mutation. |
| Side effect | Phased fan-out of subagent tasks under per-task contracts writes evidence files in disjoint scopes; adversarial and consistency critics verify. |
| Done | All workflow phases complete, the parent's shared proof run passes, and circuit breakers were honored. |

## Inputs

Required: the work item. Optional: any existing evidence ledger at the artifact path. The skill does not read conversation memory or another skill's output.

## Procedure

1. Route. If the work is a quick lookup, a single edit, an ordered plan with per-task review gates, or a flat split with no phase structure, do that work inline and stop. Only proceed when the work decomposes into parallel slices, needs independent adversarial checks, or exceeds one context window. Done when: routing decision is made; inline work is done, or the workflow proceeds.
2. Scout. Scout inline until the full work list can be named. List the files, scope the diff, find the call sites. Do not spawn workers while scouting. Done when: full work list is named.
3. Order phases. Order the workflow as phases. A phase is one wave of parallel tasks plus a barrier. Later phases consume earlier phases' evidence. Name each phase explicitly. Done when: phases are ordered and named.
4. Batch context. Carry the shared contract for the whole wave in the batch context: `# Goal` (what the wave accomplishes), `# Constraints` (rules, non-goals, permissions, verification limits), `# Contract` (shared interfaces, output shape, coordination rules). Done when: batch context is composed.
5. Per-task assignments. Each assignment is self-contained: `# Target` (exact files, symbols, or evidence surface; explicit non-goals), `# Change` (what to inspect or modify, step by step, patterns to reuse), `# Acceptance` (observable result and return packet). Workers skip formatters, linters, and project-wide tests; the parent runs shared proof once. Done when: all assignments are composed.
6. Disjoint write scopes. Every writing worker owns its paths exclusively. Shared files (manifests, configs, indexes) are edited only by the parent. If two workers must write one file, re-cut the wave before dispatching. Done when: write scopes are disjoint with no shared-file conflicts.
7. Pointers, not payloads. Workers exchange file paths and artifacts, never pasted blobs. Done when: exchange protocol is established as pointers.
8. Dispatch phase. Run the full wave. Workers execute independently. The parent stays idle until workers return. Done when: wave is dispatched and workers return.
9. Circuit breaker. Give each batch a success threshold. When a batch falls below it, stop the workflow and rediagnose instead of spending the remaining budget on a broken playbook. Done when: success threshold is set and evaluated.
10. Parent proof pass. After each wave, the parent reads returned evidence, resolves contradictions, and runs the shared proof. A wave declared done without the parent's own proof pass is a red flag. Done when: parent proof pass is run and passes, or contradictions are identified.
11. Repeat. Continue to the next phase. Later phases consume earlier phases' evidence. Done when: next phase begins or all phases are complete.
12. Close. When all phases complete and the parent's proof pass passes, the workflow is done. Done when: workflow is closed with all phases complete and proof pass passing.

### Materialize on the host

Detect the host environment and apply the matching fan-out primitive per `references/host-materialization.md` (Claude Code, oh-my-pi, or neither).

## Closing chain: Work -> Review-fix-loop -> Offensive Aggressive Simplify

1. Work first. Complete the skill's own implementation path before any closing pass.
2. Review-fix-loop with native gates. Run this skill's own review/fix gate to green: the parent proof pass per wave and the circuit breaker. Bound it: max 3 fix cycles per unit, then stop as blocked/non-converged. Never widen scope to unrelated code.
3. Offensive aggressive simplify last. Offense-first: re-derive the general case from contracts, delete old structure (no shims/aliases/flags), collapse special cases, migrate every caller in the same change, one concern per atomic commit. Ask first before removing any live consumer surface. Stay inside this skill's authority. Run the skill's verifier after each increment; revert the increment on red. Close with a simplify compression pass (dead branches, dup logic, one-use wrappers) with behavior preserved and checks green.

## Failure and recovery
| Failure class | Response |
|---|---|
| Worker returns no evidence | Diagnose whether the worker ran at all; if scope was sound, retry once with the same contract. |
| Batch falls below success threshold | Stop the workflow. Rediagnose. Do not continue to the next phase. |
| Two workers collide on one file | Stop the wave. Re-cut the partition so each file is owned by one worker. |
| Parent proof pass fails | Examine the returned evidence. Fix the root cause before the next phase. |
| Coverage cap applied | Declare what was dropped and why before acting on partial evidence. Do not silently cap. |

Partial-result rule: evidence files from successful workers in a failed wave are kept; the workflow does not delete them on failure. Non-mutation rule: the parent does not edit a worker's evidence; it reads and classifies.

## Output
The parent produces a consolidated report per phase: evidence summary, contradictions resolved, proof pass result, and next-phase readiness. The final phase output is the workflow closure report.
Pliki w paczce (5)
  • SKILL.md 6 KB
  • agents/openai.yaml 1 KB
  • references/claude-code.md 3 KB
  • references/host-materialization.md 1 KB
  • references/omp.md 6 KB

Wersje

W bazie nie ma innych skilli o nazwie workflows-driven.

Migawka bazy z 28.10.2026 · wersja paczki db2d999f3c65 (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

  • 90brand-authority
    Dba o zgodność wyników z markowym lub narzuconym stylem, z wyłączeniem zmian zdalnych, poświadczeń, publikacji, wdrożeń i działań nieodwracalnych.
  • 90context-engineering
    Odświeża długą sesję ze starym lub sprzecznym kontekstem, bez zmian zdalnych, poświadczeń i nieodwracalnych.
  • 100append-run-log
    Zapisuje zakończony przebieg agenta jako trwały, przeszukiwalny dowód w dzienniku uruchomień, z wyłączeniem zmian zdalnych, związanych z poświadczeniami, publikacją, wdrożeniem lub nieodwracalnych.
  • 100dynamic-workflows
    Orkiestruje dziesiątki lub setki subagentów ze skryptu napisanego przez Claude'a, gdy zadanie przekracza kontekst jednego agenta, jak duże audyty, migracje, badania z wielu źródeł i przeglądy równoległe.
  • 100dynamic-workflows
    Orkiestruje dziesiątki lub setki subagentów ze skryptu napisanego przez Claude'a, gdy zadanie przekracza kontekst jednego agenta, jak duże audyty, migracje, badania z wielu źródeł i przeglądy równoległe.

Uwagi i zgłoszenia

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