airlock
/airlock
Odczytuje dane z prywatnych zasobów AWS, np. baz SQL, klastrów Kubernetes, cache, tabel DynamoDB, kolejek SQS i logów, przez narzędzia MCP airlock.
- Autor
- n2jsoft-public-org ↗
- Repozytorium
- airlock ↗
- Licencja
- Apache-2.0
- Ostatnia zmiana repozytorium
- 09.11.2026
Ocena i bezpieczeństwo
- Pliki w paczce
- 3
- Skrypty
- 2
- Komentarze HTML
- 0
- Adresy URL w SKILL.md
- 0
Brak sygnałów. Repozytorium zmienione 09.11.2026, bez odwołań do wycofanych modeli i API.
Apache-2.0, otwarta: można używać i zmieniać, z zachowaniem licencji i informacji o zmianach.
- 2 skrypty w paczce. Przejrzyj je przed instalacją, bo Claude może je uruchomić.
Skrypty: agentskill.go, agentskill_test.go
Co robi i kiedy
Odczytuje dane z prywatnych zasobów AWS, np. baz SQL, klastrów Kubernetes, cache, tabel DynamoDB, kolejek SQS i logów, przez narzędzia MCP airlock.
Oryginalny opis (EN)
Reach private resources (RDS and other SQL databases, EKS clusters, ElastiCache and Docker caches, DynamoDB tables, S3 buckets, CloudWatch Logs, SQS queues and SNS topics) through the airlock MCP tools. Use when the user asks to query a database, inspect a Kubernetes cluster, read a cache key, a DynamoDB item, an S3 object, Lambda / ECS logs or an SQS queue, or to investigate data or an incident in an AWS account.
Jak działa
- Workflow with the gateway
- Rules
- Per resource
- No airlock tools?
Podgląd SKILL.md
--- name: airlock description: Reach private resources (RDS and other SQL databases, EKS clusters, ElastiCache and Docker caches, DynamoDB tables, S3 buckets, CloudWatch Logs, SQS queues and SNS topics) through the airlock MCP tools. Use when the user asks to query a database, inspect a Kubernetes cluster, read a cache key, a DynamoDB item, an S3 object, Lambda / ECS logs or an SQS queue, or to investigate data or an incident in an AWS account. --- # airlock airlock opens connections to private resources (in AWS, on direct hosts or in the local Docker) and serves MCP tools on them. Its tools come in two shapes: - **Gateway** (the airlock daemon): `list_targets`, `list_sessions`, `open_session`, `close_session`, then `db_`, `k8s_`, `cache_`, `dynamo_`, `s3_`, `logs_` and `sqs_` tools taking a `session` id. - **Session servers** (an airlock the user runs in a terminal): one MCP server per open session, whose tools carry no prefix (`query`, `list_tables`, `pod_logs`…). ## Workflow with the gateway 1. `list_sessions`: reuse a session already open on the target rather than opening another one. 2. `list_targets`: without `profile`, the saved connections and local databases (no AWS call); with `profile`, a live search of that AWS profile. Each target says whether it opens at once (`opens: auto`) or after the user approves it in the airlock dashboard (`opens: prompt`). A database server lists the `databases` seen at its last connection: to find a database by name, pass it as `query` instead of opening every server. 3. `open_session` with a short, honest `reason`: the user reads it before approving. When approval is needed, tell the user to approve the session in the airlock dashboard. 4. Call the tools of the session's family (`tools:` in `list_sessions`) with its `session` id. 5. `close_session` on the sessions you opened once done. Sessions the user opened stay open. A `.airlock.json` at the root of the user's repository lists, by stack, the targets the project uses: prefer those targets. The user opens a whole stack with `airlock up [stack]` in a terminal. ## Rules - Stay read-only unless the user asked for a change. Ask for `writes` only for a change the user asked for, and confirm the exact statement or object before running it. - Production targets deserve extra care: prefer narrow queries, never scan large tables or buckets to answer a question one targeted lookup answers. - Show the user what you ran (the query, the key, the object) along with what it returned. - Never copy secrets, passwords or tokens read from a resource into files or messages. ## Per resource Tool names below are those of a session server; the gateway prefixes them with their family (`db_query`, `k8s_pod_logs`, `cache_info`, `dynamo_scan`, `s3_list_objects`, `logs_query`). - **SQL**: explore with `list_databases`, `list_tables` and `describe_table` before writing a query. `query` runs one SELECT/WITH statement and caps the rows. `query_plan` explains a slow query (SQL Server, PostgreSQL, MySQL and MariaDB). - **Kubernetes**: `list_namespaces`, `list_resource_types`, then `list_resources` and `get_resource`; `list_events` and `pod_logs` for a failing workload. - **Cache**: `info` first; `scan_keys` with a pattern rather than reading every key. - **DynamoDB**: `describe_table` for its keys and indexes; prefer `query` (by key) over `scan`. - **S3**: `list_objects` with a prefix; `head_object` before downloading a large object. - **CloudWatch Logs**: `list_log_groups` with a prefix (`/aws/lambda/`, `/ecs/`), then `filter_events` to tail a group or `query` (Logs Insights, billed per GB scanned) over a short range first: widen it only when needed. - **SQS and SNS**: `list_queues` then `get_queue` (counts, redrive policy, DLQ); `list_topics`, `get_topic`, `get_subscription`. `peek_messages` (writes only) counts as a receive and uses up a delivery, and is refused when `maxReceiveCount` is 2 or less: never re-read a message flagged `at_limit`, and ask the user before `repost_message` or `delete_message` (served only when the user's config allows dequeuing). ## No airlock tools? - The gateway runs in the airlock daemon: the user enables it with `airlock config set mode daemon` and lists the AI clients it registers in with `airlock config set mcp.clients claude:user,codex:user` (any of claude, vscode, codex, cursor, gemini). `airlock daemon status` tells whether it runs and where it is registered. - Without the daemon, the user opens a session with `airlock` (interactive) or `airlock rds <name>`: its MCP server is registered in the project's AI client files while it runs. The client may need to reload its MCP servers to see it.
Pliki w paczce (3)
- SKILL.md 5 KB
- agentskill.go 5 KB
- agentskill_test.go 6 KB
Wersje
W bazie nie ma innych skilli o nazwie airlock.
Migawka bazy z 13.11.2026 · wersja paczki e7f62ba2a9a6 (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
- 100aws
Integruje AWS SDK w Node.js (v3) i Pythonie (boto3): S3, Lambda, DynamoDB, SQS, SNS, CloudWatch i IAM, wraz z konfiguracją, poświadczeniami i wzorcami produkcyjnymi. - 100dt-obs-aws
Monitoruje zasoby chmury AWS (EC2, RDS, Lambda, ECS i EKS, VPC, load balancery, S3, DynamoDB, SQS i SNS) pod kątem inwentarza, zgodności, planowania pojemności i oszczędności. - 95serverless-aws
Podpowiada wzorce pracy z AWS Lambda, DynamoDB, SQS i Secrets Manager w projektach serverless. - 87azure-to-aws
Planuje i wykonuje przeniesienie obciążeń z Azure do AWS: EKS, RDS, EC2, DynamoDB i ElastiCache w miejsce usług Azure. - 90aws-serverless
Pisze produkcyjne funkcje Lambda, przepływy zdarzeń przez API Gateway, SQS, SNS i DynamoDB oraz wdrożenia SAM i CDK. - 87azure-to-aws
Prowadzi migrację obciążeń z Microsoft Azure do AWS: aplikacji, baz danych, maszyn wirtualnych i kontenerów, z dopasowaniem usług.