koan-explain
/koan-explain
Wyjaśnia, jak działa aplikacja Koan na podstawie lokalnych danych, nie zmieniając przy tym plików.
- Autor
- sylin-org ↗
- Repozytorium
- koan-framework ↗
- Licencja
- Apache-2.0
- Ostatnia zmiana repozytorium
- 27.09.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 4
- Skrypty
- 0
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 27.09.2026, bez odwołań do wycofanych modeli i API.
Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.
Co robi i kiedy
Wyjaśnia, jak działa aplikacja Koan na podstawie lokalnych danych, nie zmieniając przy tym plików.
Oryginalny opis (EN)
Explain a Koan application from local evidence without changing files or state. Use when asking what an app does, how Entities, routes, jobs, AI, or MCP compose, why a provider or route was selected, what facts, health, or koan.lock.json mean, where behavior lives, or why observed runtime state or failure occurs. This skill is strictly read-only. For changes use koan; for moving an older Koan app to current Koan use koan-upgrade.
Jak działa
- Read-only boundary
- Ask the application first
- Workflow
- Interpretation rules
- Broad answer shape
- Completion check
Podgląd SKILL.md
--- name: koan-explain description: Explain a Koan application from local evidence without changing files or state. Use when asking what an app does, how Entities, routes, jobs, AI, or MCP compose, why a provider or route was selected, what facts, health, or koan.lock.json mean, where behavior lives, or why observed runtime state or failure occurs. This skill is strictly read-only. For changes use koan; for moving an older Koan app to current Koan use koan-upgrade. --- # Koan Explain Explain what the application means and why it behaves that way. Keep the answer grounded in the application, concise, and entirely read-only. ## Read-only boundary You may list, search, and read repository files; inspect existing project metadata, source, tests, build artifacts, lockfiles, logs, facts, and health output; and query an already-running documented read-only endpoint when access is already authorized. Do not edit, create, delete, format, generate, install, restore, build, start, stop, migrate, deploy, or otherwise change files, processes, configuration, data, or external state. Do not make a small fix to prove an explanation. If the request also asks for changes, complete the explanation first and hand implementation to `koan` or a framework-version migration to `koan-upgrade`. ## Ask the application first A Koan application describes itself. Prefer its own account over inference from source: | Question | Where it answers | |---|---| | Is the process alive? | `/health/live` | | Are required dependencies ready, and which one is not? | `/health/ready` | | What composed, which provider won, and why? | `/.well-known/Koan/facts` · `koan://facts` | | What is agent-visible, and under what rules? | `koan://entities` · `koan://self` | | What did references compose, and has it drifted? | `koan.lock.json` | Facts and health are redacted projections of the same runtime decisions, so quoting a provider ID or setting name from them is safe; quoting a connection string is not. These answer for **one run**. A captured response explains the run that produced it, not the repository in general. If no such output exists, say so — do not start the application to create it. ## Workflow 1. Restate the question in application terms. 2. Read repository instructions and the smallest local evidence set that can answer it. 3. Trace the relevant path: public doorway -> application owner -> capability or route -> provider -> result or failure. 4. Separate declared intent from effective runtime selection. 5. Mark material conclusions **Observed**, **Inferred**, or **Unknown**. 6. Lead with the answer, cite decisive local evidence, and name one read-only next check only when useful. Load [evidence.md](references/evidence.md) for evidence selection and uncertainty rules. Load [output-contract.md](references/output-contract.md) only when the explanation spans several owners or claims. ## Interpretation rules - Project references show available intent, not which provider won or whether the capability works. - Application source and tests show owned behavior; configuration shows declared intent. - Existing facts, health, lock, and captured runtime output show effective composition for that run. - Framework source or docs can explain a mechanism only when they match the application's resolved version. - Samples and design notes provide context, not proof of this application's behavior. - When evidence conflicts or a version match cannot be established without writing, say **Unknown**. - Redact secrets and sensitive Entity values; names of settings and providers are usually enough. For an older application, explain only behavior observed in its own evidence. Do not present old APIs as current guidance. Route requested modernization to `koan-upgrade`. ## Broad answer shape For "explain this app," cover only what matters: 1. **Purpose** - the user or operator outcome. 2. **Application vocabulary** - key Entities and operations, with their owners. 3. **Doorways and flow** - HTTP, MCP, jobs, events, storage, or other participating surfaces. 4. **Composition** - selected routes/providers and the evidence for them. 5. **Guarantees and edges** - what remains true and what is unknown or externally owned. Use a table or flow only when it makes three or more relationships easier to understand. Do not narrate every file. ## Completion check The developer should know what the app does, why its pieces participate, which conclusions are proved, and what remains unknown. The working tree and external state must be unchanged.
Pliki w paczce (4)
- SKILL.md 4 KB
- agents/openai.yaml 1 KB
- references/evidence.md 2 KB
- references/output-contract.md 1 KB
Wersje
W bazie nie ma innych skilli o nazwie koan-explain.
Migawka bazy z 13.11.2026 · wersja paczki 2b41680f5d2b (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
- 100koan-upgrade
Audytuje, planuje i migruje starszą aplikację Koan do bieżącej wersji, zachowując obserwowalne kontrakty i zatrzymując się, gdy nie da się wykazać zamiennika. - 100koan
Buduje, rozszerza, naprawia i weryfikuje aplikacje Koan: od celu biznesowego do aplikacji opartej na encjach, z doborem dostawców danych, tożsamości, wielodostępności, zadań, komunikacji, AI, MCP i testów. - 100gitly-status
Wyjaśnia, czy praca w projekcie jest zapisana i zabezpieczona, na podstawie lokalnych danych z Git. - 90koan
Pozwala trwać w paradoksie bez jego rozwiązywania: powstrzymuje odruch zwinięcia do jednej strony, trzyma dwie prawdy naraz, analizuje założenia, negację apofatyczną i etymologię. - 90explain-code
Wyjaśnia, jak działa funkcja, plik lub przepływ w kodzie, z odniesieniami do plików i linii.