Use when the user needs to generate a log collection configuration file (OpenTelemetry Collector or Elastic Beats / Filebeat) that writes logs to Alibaba Cloud Elasticsearch or self-managed Elasticsearch. Covers file logs, log API ingestion (OTLP / HTTP webhook), and Kafka log topics. Also matches natural requests such as: write a filebeat.yml, generate an otelcol config, send logs to ES, collect nginx or Java application logs to Elasticsearch, or sync Kafka logs to ES. Always output one configuration file targeting Elasticsearch; never mix technology stacks or emit non-ES outputs.
설치
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
Use when the user needs to generate a log collection configuration file (OpenTelemetry Collector or Elastic Beats / Filebeat) that writes logs to Alibaba Cloud Elasticsearch or self-managed Elasticsearch. Covers file logs, log API ingestion (OTLP / HTTP webhook), and Kafka log topics. Also matches natural requests such as: write a filebeat.yml, generate an otelcol config, send logs to ES, collect nginx or Java application logs to Elasticsearch, or sync Kafka logs to ES. Always output one configuration file targeting Elasticsearch; never mix technology stacks or emit non-ES outputs.
license
Apache-2.0
compatibility
Works with cloud-hosted or self-managed Elasticsearch. ES 8.12+ is recommended; OpenTelemetry mapping mode otel is recommended on ES 8.16+. Requires local otelcol-contrib or filebeat for configuration validation and runtime execution. This skill does not require online APIs; all references live under references/ and should be loaded on demand.
This skill generates exactly one valid configuration file that collects log data and sends it to Elasticsearch. It supports two technology stacks, and each request must choose exactly one:
OpenTelemetry Collector (otelcol-contrib) - use filelogreceiver, otlpreceiver, or kafkareceiver with elasticsearchexporter.
Elastic Beats / Filebeat - use filestream, kafka, or http_endpoint input with output.elasticsearch.
Hard Constraints
Generate only one configuration per request. Do not concatenate or batch multiple configurations.
Use exactly one technology stack. Never mix OpenTelemetry components and Filebeat sections in the same file.
Confirm before generating. Do not generate YAML until the user has answered the Step 1 technology stack confirmation question in a separate reply.
The output must be Elasticsearch. Never use Kafka, S3, Logstash-only, files, or any other non-Elasticsearch destination as the final output.
Logs only. Inputs must be log-shaped sources: files, log webhooks, or Kafka topics that contain logs. If the user asks for a clearly different non-log input, refuse and explain why.
No Syslog or Fluent forward input sources. Do not generate collector or Filebeat listener configurations for Syslog or Fluent forward protocols. If the user asks for Syslog or Fluent forward as the source, refuse and ask them to choose file, OTLP, HTTP webhook, or Kafka log ingestion instead.
Use local knowledge first. Authoritative references live in the references/ directory next to this SKILL.md. When a component option is covered locally, do not search the web; read the matching file under references/.
Keep the design minimal. Generate only fields requested by the user or required for the file to load. For common but optional features such as batching, retries, or ILM, recommend them and wait for explicit user confirmation before adding them.
Never inline secrets. Reference all secrets through ${env:VAR} for OTel or ${VAR} for Filebeat.
Local References (Prefer Over Web Search)
Before writing YAML, read the matching file under references/ to verify field names and defaults. The list below is complete. If the required option is not covered, ask the user before writing it.
Security And RAM Permissions
references/ram-policies.md - declares required_permissions for Alibaba Cloud RAM review.
08-resourcedetectionprocessor.md - host, cloud, and k8s metadata.
09-transformprocessor.md - OTTL transforms, including setting elastic.mapping.mode.
12-filestorage-extension.md - persistent offsets and queues.
Transitive references used by the components above and already localized under hard constraint #5:
13-confighttp.md - shared HTTP client/server configuration: compression algorithms gzip/zstd/snappy/zlib/deflate/lz4, compression_params.level, headers, timeouts, keep-alive, CORS, maximum body size. Used by elasticsearchexporter and otlpreceiver/http.
14-configtls.md - TLS settings: ca_file, cert_file, key_file, min_version, insecure_skip_verify, mTLS client_ca_file, TPM. Used by confighttp and configgrpc.
15-configauth.md - authentication extension wiring with auth.authenticator: <ext-name> and common client/server authentication extensions.
16-configgrpc.md - gRPC client/server settings: max_recv_msg_size_mib, keepalive, compression, balancer_name. Used by otlpreceiver/grpc.
17-exporterhelper.md - sending_queue, including persistent queues with storage, retry_on_failure, timeout, and batcher. Inherited by elasticsearchexporter.
18-ottl-overview.md - OpenTelemetry Transformation Language (OTTL) overview, statements, and contexts. Used by transformprocessor and filterprocessor.
19-ottl-functions.md - OTTL syntax and the full editor/converter catalog: set, replace_pattern, IsMatch, ParseJSON, and more.
20-ottl-log-paths.md - ottllog context paths such as log.body, log.severity_number, attributes, and SEVERITY_NUMBER_* enums.
21-stanza-operators-overview.md - operator catalog and shared fields. Used in the operators: block of filelogreceiver.
22-stanza-json_parser.md - JSON parser operator, including embedded timestamp: and severity:.
23-stanza-regex_parser.md - regex parser with named captures.
24-stanza-recombine.md - multiline recombination for stack traces and records with date prefixes.
25-stanza-timestamp.md - strptime, gotime, epoch layouts, and timezone handling.
26-stanza-severity.md - severity tokens and mapping syntax.
39-resourcedetection-detectors.md - option tables for detectors such as env, ec2, eks, aks, k8snode, kubeadm, gcp, azure, docker, openshift, and heroku; companion to 08-resourcedetectionprocessor.md.
40-auth-extensions.md - basicauth, bearertokenauth, and headers_setter extensions, wired through 15-configauth.md.
Common processors and connectors to load only when the user asks for the related capability:
27-memorylimiterprocessor.md - memory_limiter; must be first in every pipeline. Covers check_interval, limit_mib, spike_limit_mib, and soft/hard limit behavior.
28-filterprocessor.md - drop logs by OTTL conditions: log_conditions, error_mode, contexts.
29-k8sattributesprocessor.md - enrich logs with K8s pod, namespace, and node metadata: auth_type, passthrough, extract.metadata, pod_association; includes RBAC snippets.
30-redactionprocessor.md - allow-listed fields plus regex masking or hashing: allowed_keys, blocked_values, hash_function.
13-filebeat-processors-detail.md - option tables for the most common processors: add_host_metadata, add_cloud_metadata, add_kubernetes_metadata with RBAC, add_docker_metadata, decode_json_fields, dissect with tokenizer modifiers, timestamp with Go layouts, fingerprint, script, drop_event, and the field modification family: add_fields, drop_fields, rename, copy_fields, convert, truncate_fields, urldecode, replace, extract_array.
Filestream parsers to load when parsers: is needed under 02-input-filestream.md:
16-filestream-parsers.md - multiline with pattern / count / while_pattern, ndjson, container with auto / docker / cri, and include_message; includes combined stack examples.
12-internal-queue.md - queue.mem default and queue.disk settings, capacity tuning knobs.
17-setup-and-monitoring.md - setup.kibana, logging.* including file rotation and systemd journald defaults, monitoring.* internal collection, and http.* status endpoints.
Templates - assets/templates/
When a template matches the user's requested input, prefer modifying that template instead of writing from scratch:
otel-filelog-to-es.yaml
otel-kafka-to-es.yaml
otel-otlp-to-es.yaml
filebeat-filestream-to-es.yaml
filebeat-kafka-to-es.yaml
filebeat-http_endpoint-to-es.yaml
Required Workflow
Execute the following steps in order. Do not skip steps. If any required parameter is unknown at a step, ask the user explicitly at that step.
Step 1 - Choose The Technology Stack
Ask the user:
Which collection configuration do you want to generate? (1) OpenTelemetry Collector (2) Elastic Filebeat. Choose exactly one; this request will produce one configuration file.
Wait for an explicit answer in a separate user reply. Even if the initial user request names Filebeat, OpenTelemetry, filebeat.yml, otelcol, or a stack-specific input, do not treat that as the Step 1 confirmation.
Do not generate YAML in the same response that asks for the technology stack confirmation. Stop after the question above unless the current conversation already contains a direct answer to that confirmation question from an earlier turn.
Step 2 - Determine The Log Input Source
Ask for the applicable input. Allowed log inputs by stack:
If the user asks for another type, such as database input, prometheus metrics, or vmware events, refuse and explain that it is outside this skill's log collection scope.
Collect the minimum required fields:
Files: absolute glob path; whether to backfill existing content (start_at: beginning / first-start behavior); multiline pattern if needed.
OTLP: listen address and port; whether TLS or an authentication extension is required.
Kafka: brokers, topic(s), protocol_version, group_id, authentication with SASL mechanism and credentials, TLS.
Authentication - choose exactly one: basic auth with user/passwordor API key with api_key.
TLS - whether the endpoint is HTTPS, which CA file is trusted, or whether verification should be skipped with a warning.
Index / data stream:
OpenTelemetry: default is to write data streams through data_stream.dataset/namespace. For a static index, set logs_index.
Filebeat: default is filebeat-%{[agent.version]}-%{+yyyy.MM.dd} managed by ILM. Before disabling ILM or using a custom static index, ask a separate confirmation question after Step 1 has been answered. The initial request does not count as confirmation to disable ILM or use a custom index, even when it explicitly asks to turn ILM off.
OpenTelemetry only: mapping mode - confirm otel (default, recommended on ES 8.16+), ecs, none, raw, or bodymap. If the user is unsure, recommend otel for ES 8.12+; recommend ecs when the user wants flattened ECS-style fields. Wait for an explicit choice.
Step 4 - Recommend Optional Features, But Do Not Add Them Automatically
Group the following options into one question and wait for the user to choose. Add only the options the user confirms.
OpenTelemetry:
memory_limiter processor, recommended as the first processor in every pipeline to prevent collector OOM under backpressure; see 27-memorylimiterprocessor.md.
batch processor, recommended to reduce ES bulk overhead.
resourcedetection for host/cloud metadata.
k8sattributes to attach pod/namespace/node metadata when running in Kubernetes; see 29-k8sattributesprocessor.md; requires RBAC.
attributes / resource / transform for adding, deleting, or renaming attributes, OTTL transforms, and mapping mode overrides. Use resource for simple resource edits; use transform for OTTL.
filter to drop noisy logs by severity, body match, or attribute value through OTTL; see 28-filterprocessor.md.
redaction to allow-list fields and mask/hash sensitive values; see 30-redactionprocessor.md.
groupbyattrs to promote labels such as tenant_id to Resource attributes for cleaner ES routing and batching; see 32-groupbyattrsprocessor.md.
log_dedup to merge repeated INFO/heartbeat logs into one record with a count; see 33-logdedupprocessor.md.
routing connector to distribute traffic to multiple ES clusters or indexes by tenant, severity, or service; see 34-routingconnector.md.
file_storage extension plus storage: on filelog for restart persistence.
sending_queue / persistent queue on the exporter for reliability.
Choose the matching template under assets/templates/ and replace placeholders with confirmed values. Do not introduce fields the user has not approved unless they are the minimum required for the file to load.
Generation rules:
Generate YAML only after all required confirmation gates have explicit answers: technology stack, input source, Elasticsearch destination, and any sensitive behavior such as disabling ILM or using a custom index.
Reference credentials through environment variables only.
Use 2-space YAML indentation.
Write only service.pipelines.logs; do not write traces or metrics pipelines unless the user asks for them.
Verify that every component referenced under service: is declared at the top level.
Explain unusual choices with one short comment line; do not add long explanations.
Step 6 - Self-Check Before Returning
Mentally run this checklist. If any item fails, refuse to output the configuration:
Single technology stack: OTel or Filebeat, never both.
The Step 1 technology stack confirmation was answered in a separate user reply, or was already answered earlier in the conversation.
Output points only to Elasticsearch.
Input is log-shaped.
Every component ID in service.pipelines.logs for OTel is declared at the top level, or output.elasticsearch is the only Filebeat output.
Exactly one of endpoint / endpoints / cloudid is used in the OTel ES exporter.
Exactly one authentication method is used: basic or api_key, never both.
No inline secrets; all secrets are environment-variable references.
No fields the user did not request, except the minimum fields required for loading.
Filebeat setup.ilm.enabled: false, output.elasticsearch.index, or indices: appears only after a separate confirmation for ILM/custom index behavior.
YAML is parseable: no tabs, consistent indentation, balanced lists.
If start_at: beginning is not enabled, comment on the impact: existing file content will not be backfilled.
Then return:
The complete YAML, ready to save as otelcol-config.yaml or filebeat.yml.
A three-point "how to run / validate" footer:
OTel: otelcol-contrib validate --config=otelcol-config.yaml, then otelcol-contrib --config=otelcol-config.yaml.
Filebeat: filebeat test config -c filebeat.yml, filebeat test output, and filebeat -e -c filebeat.yml.
Confirmation Policy
When unsure, ask. Use numbered question blocks; do not silently choose. Specifically:
The initial user request never counts as the Step 1 technology stack confirmation. Only a later direct answer to the Step 1 question counts.
Do not proceed past Step 1 in the same assistant response that asks for the technology stack.
Do not pick an OTel ES exporter mapping mode when the ES version is unknown.
Do not enable start_at: beginning without confirmation, because it changes which lines are sent.
Do not disable ILM or set a custom index without a separate confirmation. The initial request does not count as confirmation to disable ILM or use a custom index, even when it explicitly asks to turn ILM off.
Do not choose a TLS verification mode, full or none, without confirmation.
Refusal Cases
Refuse these requests directly:
Output to any destination other than Elasticsearch. Logstash-only is also refused; users may run Logstash separately, but this skill targets ES directly.
Mix both technology stacks in one file.
Generate multiple configurations in a single request.
Use inputs unrelated to logs, such as raw metrics scraping or span generation.
Use Syslog or Fluent forward as the log input source.
Common Mistakes
Mistake
Fix
Inline a plaintext password in the file
Use ${env:ELASTIC_PASSWORD} for OTel or ${ELASTIC_PASSWORD} for Filebeat.
Default to start_at: end when the user wants backfill
Ask first; if the user wants existing content collected, set beginning.
Generate YAML immediately because the first prompt already says "Filebeat"
Ask the Step 1 technology stack question first, wait for a separate reply, and stop before generating YAML.
Add traces / metrics pipelines without being asked
Output only a logs pipeline unless the user requests otherwise.
Forget to declare a component referenced in service.pipelines
Always cross-check declared components against referenced components.
Blindly use mapping.mode: ecs
The default is otel; switch only when the user asks.
Set setup.ilm.enabled: false because the initial prompt asked for a custom index
Ask a separate ILM/custom-index confirmation after the stack is confirmed.
Use Filebeat ILM while also setting index:
Either confirm disabling ILM for the custom index or omit the custom index; choose one.
