使用方法
- 复制安装命令,在自己的智能体环境安装前先检查源码包。
- 从原始文档核对所需工具、凭据和支持的宿主。
- 说明输入材料与期望交付物,并对照来源检查输出。
npx skills add https://github.com/elastic/agent-skills --skill kibana-cases开始前需了解
- 安装不包含模型订阅、API 额度或访问文件的权限。
- 仓库星数和安装量反映活跃情况,不等于独立质量或安全测试。
按源仓库内容呈现,保留标题、案例、代码、表格、链接以及原文引用的演示图片。
Kibana Cases
Create, search, update, and enrich Elastic cases across Security, Observability, and Stack Management. You have full read and write access — never claim read-only access. Cases can be assigned to users, linked to alerts and events, enriched with observables, and pushed to external incident-management systems via connectors.
<!-- begin-partial: preamble -->
Environment Configuration
This skill executes Elasticsearch operations through the elastic CLI. If the `elastic` CLI is not installed, tell the user what it is needed for. Do not guess credentials, call the HTTP API directly, or attempt other workarounds.
This skill references operations in HTTP-shorthand form (e.g., GET /, GET /_cat/indices, GET /{index}/_mapping, GET /{index}/_settings/index.mode, POST /_query). The Operations table at the end of this document maps each shorthand to the equivalent elastic CLI command — always use the CLI rather than calling the HTTP API directly.
<!-- end-partial: preamble -->
Solution context — highest-priority rule
Cases are partitioned across securitySolution (Security), observability (Observability), and cases (Stack Management). A user works in exactly one. Operating in the wrong solution — or fanning out across all three — is always a bug.
Before any cases operation, the solution `owner` must be set, sourced ONLY from the user naming it in plain language ("in Security", "obs cases", "Stack Management", etc.) in the current or an earlier message. Topic-keyword inference does not count. If the owner is unset, your only valid action is to ask, verbatim:
"Which Elastic app is this in — Security, Observability, or Stack Management?"
Once set: pass it as owner on every create and as the owner filter on every _find; treat it as fixed unless the user explicitly switches. Never:
- Call
_find(or any read) with multipleownervalues to "cover all three" — this is the most common failure. - Silently fall back to another owner on empty results — say "No matching cases in
<Solution>" and stop. - Skip this for case-id operations — an id resolves a record, but the solution context still confirms intent.
Examples:
- "find the case about the failing payment service" → ask first.
- "investigating a Suricata alert in Security — find related cases" →
securitySolution(named directly). - (turn 2, Security established on turn 1) "show me the open ones" → reuse, don't re-ask.
- "close case abc-123" with no prior context → ask first.
Solution profiles
| Solution | Typical triggers | Observables (IOCs) | critical severity means |
|---|---|---|---|
securitySolution | SIEM detection rules, threat hunts, manual triage | Primary use case — proactively track IPs, domains, file hashes, URLs | active attack in progress |
observability | APM errors, SLO violations, metric/log anomalies | Rarely relevant — do not proactively suggest | complete service outage |
cases | General-purpose, no domain assumptions | Rarely relevant | — |
Domain notes: status flows open → in-progress → closed. assignees are user profile UIDs, not usernames. severity is one of low, medium, high, critical.
Process
- Establish solution context, then classify the task. Confirm the
ownerper the rule above. Decide whether the
user needs to create a case, find/list cases (read-only), get one case, comment on or attach alerts/events to a case, add observables, or update one or more cases. If the user only asks to show, list, or search, treat the request as read-only — do not create, comment, attach, or update anything.
- Check for an existing case before creating one. When triaging, first search so you do not create duplicates. Call
GET kbn:/api/cases/_find filtered to the owner, or GET kbn:/api/cases/alerts/{alertId} to see whether a specific alert is already attached to a case.
- By text:
search=<hostname or keyword>searches title, description, and comments. Addstatus=opento limit
to active cases.
- By tag:
tags=<tag>is exact-match only — use it to correlate on a host, agent, or rule.
Report the exact total count and each matching case verbatim. Reuse an existing case when the same entity or rule is already tracked; otherwise continue to create.
- For create tasks, build the payload. Call
POST kbn:/api/caseswithtitle,description,tags,severity,
the resolved owner, a connector (use the none connector unless the user asks to wire an external system), and settings.syncAlerts (default true). Confirm the create with the user before writing, then report the returned case id verbatim.
{
"title": "Malicious DLL sideloading on host1",
"description": "Crypto clipper malware detected via DLL sideloading...",
"tags": ["classification:malicious", "mitre:T1574.002"],
"severity": "critical",
"owner": "securitySolution",
"connector": { "id": "none", "name": "none", "type": ".none", "fields": null },
"settings": { "syncAlerts": true }
}To apply a case template, read the available templates from GET kbn:/api/cases/configure, then merge the template's fields (including any customFields) into the create payload.
- Attach alerts and events. Attach each alert as an
alertcomment viaPOST kbn:/api/cases/{caseId}/comments
with type: "alert", alertId, index (the alert's data-stream index), the owner, and a rule object. rule.id and rule.name are required — pass unknown for either when not known, or the request returns a 400. Attach one at a time and space repeated calls a few seconds apart to avoid rate limiting (HTTP 429).
{
"type": "alert",
"alertId": "<alert_doc_id>",
"index": ".ds-.alerts-security.alerts-default-2025.12.01-000013",
"rule": { "id": "<rule_id>", "name": "Malicious Behavior Detection Alert" },
"owner": "securitySolution"
}- Add investigation notes. Record findings with
POST kbn:/api/cases/{caseId}/commentsusingtype: "user", a
comment string, and the owner.
- Add observables (Security). For Security cases, proactively offer to track IOCs (IPs, domains, file hashes, URLs)
with POST kbn:/api/cases/{caseId}/observables. Update or remove them with the matching PATCH/DELETE observable operations. Do not proactively suggest observables for Observability or Stack Management cases.
- For update tasks, read then patch — and batch. Cases use optimistic concurrency, so you must send the current
version. Call GET kbn:/api/cases/{caseId} to read the latest version, then PATCH kbn:/api/cases with a cases array. Updating two or more cases in one request is a single PATCH with multiple array entries — never multiple sequential updates. When changing tags, merge with existing tags unless the user asks to replace them. Confirm the update with the user before writing.
{ "cases": [{ "id": "<case_id>", "version": "<case_version>", "status": "closed", "severity": "low" }] }On a version conflict, re-fetch the affected case and retry with the fresh version. Assignee-only changes go through the same PATCH with an assignees array.
- Report faithfully. Copy case ids, titles, tags, severities, and counts exactly as returned by the API. Do not
abbreviate ids, truncate titles, invent details, or round numbers. When the API returns zero results, state that explicitly.
Reporting list and find results
- State the exact
totalcount from the JSON response (e.g., "There are 12 open cases total"). - Present each case as a compact one-line entry:
<title> | <severity> | <case_id_short> | <created_at>. Copy the
exact title verbatim — do not rephrase, abbreviate, or summarize.
- If the user asked for N cases, present exactly N entries (or fewer if fewer exist). Do not add columns (alert count,
description, status) unless the user asked for them.
- Omit fields that are null or missing. Do not add analysis or commentary after the list.
Comments and attachments
Cases read operations return metadata only — comments and attachments are never included. To analyze discussion or list attached alerts, call GET kbn:/api/cases/{caseId}/comments/_find and filter to type === "user" for comments. Only fetch when the user explicitly asks for a case summary that includes discussion, a quote of comments, or the attached alerts/events for a specific case. Never preemptively fetch for cases in a list.
Examples
Create a case after triage (Security)
User: "In Security, create a case for the phishing alert I triaged with severity high."
- Owner is
securitySolution(named). Search first withGET kbn:/api/cases/_find(or
GET kbn:/api/cases/alerts/{alertId}) to avoid a duplicate.
POST kbn:/api/caseswithowner: "securitySolution", thenoneconnector, andseverity: "high".- Attach the alert with
POST kbn:/api/cases/{caseId}/comments(type: "alert",rule.id/rule.nameset). - Report the returned case
idverbatim.
Find open cases (read-only)
User: "Show me open Observability cases about the checkout service."
- Owner is
observability(named).GET kbn:/api/cases/_findwithsearch=checkout,status=open,
owner=observability, sortField=createdAt, sortOrder=desc.
- Report the exact
totaland each title verbatim. Do not mutate anything.
Close several cases at once
User: "Close cases abc-123, def-456, and ghi-789."
GETeach case for its currentversion.- One
PATCH kbn:/api/caseswith all three entries (status: "closed") — not three separate calls.
Ambiguous — ask first
User: "Find the case about the failing payment service."
- No solution named → ask verbatim: "Which Elastic app is this in — Security, Observability, or Stack Management?"
Guidelines
- Report only tool output — do not invent ids, hostnames, IPs, or details not present in the response.
- Confirm write actions (create, comment, attach, update, delete) with the user before executing them.
- The solution
owneris fixed once established — never query or write across multiple owners in one request. _findwithsearchmay return 500 errors on Serverless; prefertagsfiltering or a plain list there.tagsfiltering is exact-match only — partial matches do not work.- Attach alerts one at a time via the public comments endpoint; the bulk attachment endpoint is internal (see
- For non-default Kibana Spaces, prefix paths with
kbn:/s/<space_id>/api/cases/....
Common pitfalls
- Missing solution context — creating or searching without a confirmed
owner. - Fanning out across owners — passing multiple
ownervalues to "cover all three". - Sequential updates — issuing N
PATCHcalls instead of one bulkPATCHfor a multi-case change. - Stale version —
PATCHwithout a freshGETreturns a version conflict. - Read-only violations — mutating cases when the user only asked to list or search.
- Over-fetching — pulling comments/attachments for every case in a list instead of on request.
References
- references/kibana-cases-api.md — Endpoints, request/response shapes, query params
- Kibana Cases API
Operations
| HTTP API (shorthand) | elastic CLI command |
|---|---|
POST kbn:/api/cases | elastic kb cases create-case --input-file '<case.json>' |
GET kbn:/api/cases/_find | elastic kb cases find-cases ¹ |
GET kbn:/api/cases/{caseId} | elastic kb cases get-case --case-id '<id>' |
PATCH kbn:/api/cases | elastic kb cases update-case --input-file '<cases.json>' |
DELETE kbn:/api/cases | elastic kb cases delete-case --yes ² |
POST kbn:/api/cases/{caseId}/comments | elastic kb cases add-case-comment --case-id '<id>' --input-file '<attachment.json>' |
GET kbn:/api/cases/{caseId}/comments/_find | elastic kb cases find-case-comments --case-id '<id>' |
GET kbn:/api/cases/{caseId}/alerts | elastic kb cases get-case-alerts --case-id '<id>' |
GET kbn:/api/cases/alerts/{alertId} | elastic kb cases get-cases-by-alert --alert-id '<id>' [--owner '<owner>'] |
GET kbn:/api/cases/configure | elastic kb cases get-case-configuration |
POST kbn:/api/cases/{caseId}/observables | (no CLI binding — use HTTP shorthand) |
PATCH kbn:/api/cases/{caseId}/observables/{obsId} | (no CLI binding — use HTTP shorthand) |
DELETE kbn:/api/cases/{caseId}/observables/{obsId} | (no CLI binding — use HTTP shorthand) |
¹ find-cases exposes no filter flags in the current CLI release; for owner, status, tags, search, and pagination, run GET kbn:/api/cases/_find with query params via the HTTP transport directly.
² delete-case exposes no --ids flag in the current CLI release; to target specific cases, run DELETE kbn:/api/cases?ids=<json-array> via the HTTP transport directly.
Internal-only endpoints (no public binding). Bulk get (POST /internal/cases/_bulk_get), find-similar (POST /internal/cases/{caseId}/_similar), and bulk attachment (POST /internal/cases/{caseId}/attachments/_bulk_create) are internal — use the public per-item operations above instead. Assignee changes and create_from_template have no dedicated endpoints: set assignees through PATCH /api/cases, and apply templates by merging GET /api/cases/configure output into the create payload.