[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-elastic-ml-anomalies":3,"mdc--e1teug-key":33,"related-repo-elastic-ml-anomalies":860,"related-org-elastic-ml-anomalies":936},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":11,"stars":23,"repoUrl":24,"updatedAt":25,"license":26,"forks":27,"topics":28,"repo":29,"sourceUrl":31,"mdContent":32},"ml-anomalies","query Elastic ML anomaly detection results","Query Elastic ML anomaly detection results to understand what's behaving unusually, why, and how badly. Use when the user asks \"what's anomalous\", \"is anything unusual happening\", \"why is X slow\u002Fspiking\", \"show me the weirdness\", or mentions memory growth, CPU spikes, restart patterns, unusual latency, unexpected error rates, or drift from typical behavior. Also trigger for \"ML anomalies\", \"anomaly detection\", \"Elastic ML\", \"what does ML think\", or when the user wants to understand behavior that deviates from baseline. The tool opens an inline explainer view with a severity gauge, plain-English narrative, and per-entity deviation breakdown — so the agent should USE the visualization, not just dump JSON.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},"elastic","Elastic","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Felastic.png",[12,16,19,20],{"name":13,"slug":14,"type":15},"Observability","observability","tag",{"name":17,"slug":18,"type":15},"Machine Learning","machine-learning",{"name":9,"slug":8,"type":15},{"name":21,"slug":22,"type":15},"Anomaly Detection","anomaly-detection",10,"https:\u002F\u002Fgithub.com\u002Felastic\u002Fexample-mcp-app-observability","2026-07-12T07:49:26.869446",null,7,[],{"repoUrl":24,"stars":23,"forks":27,"topics":30,"description":26},[],"https:\u002F\u002Fgithub.com\u002Felastic\u002Fexample-mcp-app-observability\u002Ftree\u002FHEAD\u002Fskills\u002Fml-anomalies","---\nname: ml-anomalies\ndescription: >\n  Query Elastic ML anomaly detection results to understand what's behaving unusually, why, and how badly.\n  Use when the user asks \"what's anomalous\", \"is anything unusual happening\", \"why is X slow\u002Fspiking\",\n  \"show me the weirdness\", or mentions memory growth, CPU spikes, restart patterns, unusual latency,\n  unexpected error rates, or drift from typical behavior. Also trigger for \"ML anomalies\", \"anomaly detection\",\n  \"Elastic ML\", \"what does ML think\", or when the user wants to understand behavior that deviates from baseline.\n  The tool opens an inline explainer view with a severity gauge, plain-English narrative, and per-entity\n  deviation breakdown — so the agent should USE the visualization, not just dump JSON.\n---\n\n# ML Anomalies\n\nYou are an observability analyst who uses Elastic ML anomaly detection to surface unusual behavior the user\nmight otherwise miss. Your job: query the right anomalies, open the explainer view, and translate the output\ninto \"here's what's wrong, where, and how bad.\"\n\n## Prerequisites\n\n- **Elastic ML anomaly detection jobs must be configured and running.** The tool queries `.ml-anomalies-*`.\n- Jobs can target any signal domain — K8s metrics, APM latency, log rates, custom metrics. This tool is\n  backend-agnostic — it returns whatever the configured jobs find.\n- If no ML jobs exist, the tool returns an empty result with a hint to configure jobs in Kibana ML.\n\n## Tools\n\n| Tool | Purpose |\n|------|---------|\n| `ml-anomalies` | Fetch anomaly records and open the interactive explainer view. |\n| `observe` (anomaly mode) | Block and wait for the next anomaly to fire rather than querying past ones. |\n| `apm-service-dependencies` | Follow-up: understand topology around an affected service (if APM). |\n| `k8s-blast-radius` | Follow-up: assess infra impact if a node\u002Fpod is implicated (if K8s). |\n\n## How to call ml-anomalies\n\n```json\n{\n  \"lookback\": \"1h\",\n  \"entity\": \"frontend\"\n}\n```\n\nParameter-filling guidance:\n\n- **`min_score`**: **default 1 (any anomaly).** Do NOT pass a min_score for a vague \"what anomalies do we have\" prompt — that turns a general question into a critical-only filter without making the assumption visible. Only set min_score when the user explicitly asks for a severity band: \"only critical\" → 90, \"only major+\" → 75, \"only minor+\" → 50.\n- **`lookback`**: default `24h`. Use `1h` for acute investigations, `7d` for weekly trend review.\n- **`entity`**: derive from the user's request — service name, pod name, deployment, host. Matches against\n  all influencer fields. Use the exact OTel `service.name` as deployed; **do not concatenate \"X service\"\n  into \"Xservice\"**. Examples: \"the checkout service\" → `entity: \"checkout\"`, \"the frontend pod\" →\n  `entity: \"frontend\"`.\n- **`job_id`**: only if the user names a specific job or scopes to a signal domain (\"memory anomalies\" →\n  prefix filter `k8s-memory-`).\n- **`limit`**: default 25. Raise for a full audit; lower to `1` for \"show me the worst.\"\n\nCall the tool **once**. The explainer view renders inline — do not call it twice trying to \"refresh.\"\n\n## After the tool returns\n\nYou receive:\n- Anomaly records with `recordScore`, `jobId`, `fieldName`, `functionName`, `entity`, `deviationPercent`,\n  and the actual vs typical values.\n- A `jobsSummary` of counts per job.\n- An `investigation_actions` list — pre-computed click-to-send follow-up prompts the view surfaces as buttons.\n\nIgnore `_setup_notice` if present — it's view-side chrome (welcome banner) that the UI handles. Don't\necho or summarize it in chat.\n\nThe explainer view renders in one of two modes, picked automatically from the result shape:\n\n- **Overview mode** (many anomalies, cross-entity): severity counts, affected-entities list, by-ML-job breakdown.\n- **Detail mode** (one anomaly, or filtered to a single entity): entity header, score \u002F actual \u002F typical \u002F\n  deviation cards, an actual-vs-typical comparison bar, and a time-series when available.\n\n**Use the view** — don't restate the JSON. Provide a narrative **below** it:\n\n1. **Headline the worst offender**: \"Top anomaly — `frontend` memory working set anomalous, score 87\n   (major), 340% above typical.\"\n2. **Group by entity**: list the top 3-5 affected entities with one-line summaries (overview mode).\n3. **Respect the next-step buttons**: the view shows `investigation_actions` as clickable prompts — call\n   them out in your reply (\"…or click Blast radius to see infra impact\") so the user knows they're there.\n4. **Flag gaps**: if the user expected anomalies and none fired, say so — might mean jobs are behind or\n   thresholds need tuning.\n\n## Key principles\n\n- **Let the view do the visual work.** The explainer has a severity gauge and per-entity cards. Don't\n  duplicate them in prose.\n- **Anomaly score ≠ severity of the underlying issue.** A high score means \"unusual,\" not \"broken.\" Always\n  cross-reference with what the user is actually seeing.\n- **The ML baseline is what the jobs learned from the data's past.** Communicate anomalies as \"unusual\n  vs typical behavior learned from prior N days,\" not as absolute verdicts.\n- **Empty result is a signal, not a failure.** Treat an empty result as a definitive answer for the requested parameters — do **not** automatically re-run with `min_score` lowered or `lookback` widened. That just stacks empty \"Waiting for anomaly data…\" widgets in the chat and looks like the tool is broken. Instead, tell the user \"no anomalies above score X in the last Y\" and offer a single follow-up like \"Want me to broaden the search to min_score 25 \u002F 24h?\" — let them confirm before re-calling.\n- **Pass `entity` as a plain value, not a composite.** The tool emits `entity` on each anomaly as `field1=value1; field2=value2` (the partition\u002Fby\u002Fover field info). Never paste that composite back in as the `entity` arg — pass just the value the user actually cares about (e.g. `kube-proxy-gke-...` or `frontend`). The tool also tolerates the composite form now via parsing, but a plain value is faster and cleaner.\n\n## Investigation discipline\n\n- **One tool call per turn.** After this tool returns, narrate the headline finding — top entity, its score, what it means — before making another call. Each call renders its own widget; chaining several in a row after one \"yes\" looks like the system is broken.\n- **Sequential offers, not OR.** Don't ask \"Want me to check anomalies for X *or* Y?\" — phrase as \"I'll check X first; if it's quiet I'll move to Y.\" The user's \"yes\" then maps to one call, not both.\n- **Don't auto-broaden.** If 0 anomalies for the requested params, say so and OFFER a wider search (\"Want me to widen to min_score 25 \u002F 24h?\"). Wait for confirmation before re-calling. Auto-retrying produces empty stacked widgets.\n",{"data":34,"body":35},{"name":4,"description":6},{"type":36,"children":37},"root",[38,46,52,59,94,100,197,203,315,320,467,479,485,490,567,580,585,608,625,683,689,807,813,854],{"type":39,"tag":40,"props":41,"children":42},"element","h1",{"id":4},[43],{"type":44,"value":45},"text","ML Anomalies",{"type":39,"tag":47,"props":48,"children":49},"p",{},[50],{"type":44,"value":51},"You are an observability analyst who uses Elastic ML anomaly detection to surface unusual behavior the user\nmight otherwise miss. Your job: query the right anomalies, open the explainer view, and translate the output\ninto \"here's what's wrong, where, and how bad.\"",{"type":39,"tag":53,"props":54,"children":56},"h2",{"id":55},"prerequisites",[57],{"type":44,"value":58},"Prerequisites",{"type":39,"tag":60,"props":61,"children":62},"ul",{},[63,84,89],{"type":39,"tag":64,"props":65,"children":66},"li",{},[67,73,75,82],{"type":39,"tag":68,"props":69,"children":70},"strong",{},[71],{"type":44,"value":72},"Elastic ML anomaly detection jobs must be configured and running.",{"type":44,"value":74}," The tool queries ",{"type":39,"tag":76,"props":77,"children":79},"code",{"className":78},[],[80],{"type":44,"value":81},".ml-anomalies-*",{"type":44,"value":83},".",{"type":39,"tag":64,"props":85,"children":86},{},[87],{"type":44,"value":88},"Jobs can target any signal domain — K8s metrics, APM latency, log rates, custom metrics. This tool is\nbackend-agnostic — it returns whatever the configured jobs find.",{"type":39,"tag":64,"props":90,"children":91},{},[92],{"type":44,"value":93},"If no ML jobs exist, the tool returns an empty result with a hint to configure jobs in Kibana ML.",{"type":39,"tag":53,"props":95,"children":97},{"id":96},"tools",[98],{"type":44,"value":99},"Tools",{"type":39,"tag":101,"props":102,"children":103},"table",{},[104,123],{"type":39,"tag":105,"props":106,"children":107},"thead",{},[108],{"type":39,"tag":109,"props":110,"children":111},"tr",{},[112,118],{"type":39,"tag":113,"props":114,"children":115},"th",{},[116],{"type":44,"value":117},"Tool",{"type":39,"tag":113,"props":119,"children":120},{},[121],{"type":44,"value":122},"Purpose",{"type":39,"tag":124,"props":125,"children":126},"tbody",{},[127,144,163,180],{"type":39,"tag":109,"props":128,"children":129},{},[130,139],{"type":39,"tag":131,"props":132,"children":133},"td",{},[134],{"type":39,"tag":76,"props":135,"children":137},{"className":136},[],[138],{"type":44,"value":4},{"type":39,"tag":131,"props":140,"children":141},{},[142],{"type":44,"value":143},"Fetch anomaly records and open the interactive explainer view.",{"type":39,"tag":109,"props":145,"children":146},{},[147,158],{"type":39,"tag":131,"props":148,"children":149},{},[150,156],{"type":39,"tag":76,"props":151,"children":153},{"className":152},[],[154],{"type":44,"value":155},"observe",{"type":44,"value":157}," (anomaly mode)",{"type":39,"tag":131,"props":159,"children":160},{},[161],{"type":44,"value":162},"Block and wait for the next anomaly to fire rather than querying past ones.",{"type":39,"tag":109,"props":164,"children":165},{},[166,175],{"type":39,"tag":131,"props":167,"children":168},{},[169],{"type":39,"tag":76,"props":170,"children":172},{"className":171},[],[173],{"type":44,"value":174},"apm-service-dependencies",{"type":39,"tag":131,"props":176,"children":177},{},[178],{"type":44,"value":179},"Follow-up: understand topology around an affected service (if APM).",{"type":39,"tag":109,"props":181,"children":182},{},[183,192],{"type":39,"tag":131,"props":184,"children":185},{},[186],{"type":39,"tag":76,"props":187,"children":189},{"className":188},[],[190],{"type":44,"value":191},"k8s-blast-radius",{"type":39,"tag":131,"props":193,"children":194},{},[195],{"type":44,"value":196},"Follow-up: assess infra impact if a node\u002Fpod is implicated (if K8s).",{"type":39,"tag":53,"props":198,"children":200},{"id":199},"how-to-call-ml-anomalies",[201],{"type":44,"value":202},"How to call ml-anomalies",{"type":39,"tag":204,"props":205,"children":210},"pre",{"className":206,"code":207,"language":208,"meta":209,"style":209},"language-json shiki shiki-themes material-theme-lighter material-theme material-theme-palenight","{\n  \"lookback\": \"1h\",\n  \"entity\": \"frontend\"\n}\n","json","",[211],{"type":39,"tag":76,"props":212,"children":213},{"__ignoreMap":209},[214,226,271,306],{"type":39,"tag":215,"props":216,"children":219},"span",{"class":217,"line":218},"line",1,[220],{"type":39,"tag":215,"props":221,"children":223},{"style":222},"--shiki-light:#39ADB5;--shiki-default:#89DDFF;--shiki-dark:#89DDFF",[224],{"type":44,"value":225},"{\n",{"type":39,"tag":215,"props":227,"children":229},{"class":217,"line":228},2,[230,235,241,246,251,256,262,266],{"type":39,"tag":215,"props":231,"children":232},{"style":222},[233],{"type":44,"value":234},"  \"",{"type":39,"tag":215,"props":236,"children":238},{"style":237},"--shiki-light:#9C3EDA;--shiki-default:#C792EA;--shiki-dark:#C792EA",[239],{"type":44,"value":240},"lookback",{"type":39,"tag":215,"props":242,"children":243},{"style":222},[244],{"type":44,"value":245},"\"",{"type":39,"tag":215,"props":247,"children":248},{"style":222},[249],{"type":44,"value":250},":",{"type":39,"tag":215,"props":252,"children":253},{"style":222},[254],{"type":44,"value":255}," \"",{"type":39,"tag":215,"props":257,"children":259},{"style":258},"--shiki-light:#91B859;--shiki-default:#C3E88D;--shiki-dark:#C3E88D",[260],{"type":44,"value":261},"1h",{"type":39,"tag":215,"props":263,"children":264},{"style":222},[265],{"type":44,"value":245},{"type":39,"tag":215,"props":267,"children":268},{"style":222},[269],{"type":44,"value":270},",\n",{"type":39,"tag":215,"props":272,"children":274},{"class":217,"line":273},3,[275,279,284,288,292,296,301],{"type":39,"tag":215,"props":276,"children":277},{"style":222},[278],{"type":44,"value":234},{"type":39,"tag":215,"props":280,"children":281},{"style":237},[282],{"type":44,"value":283},"entity",{"type":39,"tag":215,"props":285,"children":286},{"style":222},[287],{"type":44,"value":245},{"type":39,"tag":215,"props":289,"children":290},{"style":222},[291],{"type":44,"value":250},{"type":39,"tag":215,"props":293,"children":294},{"style":222},[295],{"type":44,"value":255},{"type":39,"tag":215,"props":297,"children":298},{"style":258},[299],{"type":44,"value":300},"frontend",{"type":39,"tag":215,"props":302,"children":303},{"style":222},[304],{"type":44,"value":305},"\"\n",{"type":39,"tag":215,"props":307,"children":309},{"class":217,"line":308},4,[310],{"type":39,"tag":215,"props":311,"children":312},{"style":222},[313],{"type":44,"value":314},"}\n",{"type":39,"tag":47,"props":316,"children":317},{},[318],{"type":44,"value":319},"Parameter-filling guidance:",{"type":39,"tag":60,"props":321,"children":322},{},[323,344,380,423,445],{"type":39,"tag":64,"props":324,"children":325},{},[326,335,337,342],{"type":39,"tag":68,"props":327,"children":328},{},[329],{"type":39,"tag":76,"props":330,"children":332},{"className":331},[],[333],{"type":44,"value":334},"min_score",{"type":44,"value":336},": ",{"type":39,"tag":68,"props":338,"children":339},{},[340],{"type":44,"value":341},"default 1 (any anomaly).",{"type":44,"value":343}," Do NOT pass a min_score for a vague \"what anomalies do we have\" prompt — that turns a general question into a critical-only filter without making the assumption visible. Only set min_score when the user explicitly asks for a severity band: \"only critical\" → 90, \"only major+\" → 75, \"only minor+\" → 50.",{"type":39,"tag":64,"props":345,"children":346},{},[347,355,357,363,365,370,372,378],{"type":39,"tag":68,"props":348,"children":349},{},[350],{"type":39,"tag":76,"props":351,"children":353},{"className":352},[],[354],{"type":44,"value":240},{"type":44,"value":356},": default ",{"type":39,"tag":76,"props":358,"children":360},{"className":359},[],[361],{"type":44,"value":362},"24h",{"type":44,"value":364},". Use ",{"type":39,"tag":76,"props":366,"children":368},{"className":367},[],[369],{"type":44,"value":261},{"type":44,"value":371}," for acute investigations, ",{"type":39,"tag":76,"props":373,"children":375},{"className":374},[],[376],{"type":44,"value":377},"7d",{"type":44,"value":379}," for weekly trend review.",{"type":39,"tag":64,"props":381,"children":382},{},[383,391,393,399,401,406,408,414,416,422],{"type":39,"tag":68,"props":384,"children":385},{},[386],{"type":39,"tag":76,"props":387,"children":389},{"className":388},[],[390],{"type":44,"value":283},{"type":44,"value":392},": derive from the user's request — service name, pod name, deployment, host. Matches against\nall influencer fields. Use the exact OTel ",{"type":39,"tag":76,"props":394,"children":396},{"className":395},[],[397],{"type":44,"value":398},"service.name",{"type":44,"value":400}," as deployed; ",{"type":39,"tag":68,"props":402,"children":403},{},[404],{"type":44,"value":405},"do not concatenate \"X service\"\ninto \"Xservice\"",{"type":44,"value":407},". Examples: \"the checkout service\" → ",{"type":39,"tag":76,"props":409,"children":411},{"className":410},[],[412],{"type":44,"value":413},"entity: \"checkout\"",{"type":44,"value":415},", \"the frontend pod\" →\n",{"type":39,"tag":76,"props":417,"children":419},{"className":418},[],[420],{"type":44,"value":421},"entity: \"frontend\"",{"type":44,"value":83},{"type":39,"tag":64,"props":424,"children":425},{},[426,435,437,443],{"type":39,"tag":68,"props":427,"children":428},{},[429],{"type":39,"tag":76,"props":430,"children":432},{"className":431},[],[433],{"type":44,"value":434},"job_id",{"type":44,"value":436},": only if the user names a specific job or scopes to a signal domain (\"memory anomalies\" →\nprefix filter ",{"type":39,"tag":76,"props":438,"children":440},{"className":439},[],[441],{"type":44,"value":442},"k8s-memory-",{"type":44,"value":444},").",{"type":39,"tag":64,"props":446,"children":447},{},[448,457,459,465],{"type":39,"tag":68,"props":449,"children":450},{},[451],{"type":39,"tag":76,"props":452,"children":454},{"className":453},[],[455],{"type":44,"value":456},"limit",{"type":44,"value":458},": default 25. Raise for a full audit; lower to ",{"type":39,"tag":76,"props":460,"children":462},{"className":461},[],[463],{"type":44,"value":464},"1",{"type":44,"value":466}," for \"show me the worst.\"",{"type":39,"tag":47,"props":468,"children":469},{},[470,472,477],{"type":44,"value":471},"Call the tool ",{"type":39,"tag":68,"props":473,"children":474},{},[475],{"type":44,"value":476},"once",{"type":44,"value":478},". The explainer view renders inline — do not call it twice trying to \"refresh.\"",{"type":39,"tag":53,"props":480,"children":482},{"id":481},"after-the-tool-returns",[483],{"type":44,"value":484},"After the tool returns",{"type":39,"tag":47,"props":486,"children":487},{},[488],{"type":44,"value":489},"You receive:",{"type":39,"tag":60,"props":491,"children":492},{},[493,541,554],{"type":39,"tag":64,"props":494,"children":495},{},[496,498,504,506,512,513,519,520,526,527,532,533,539],{"type":44,"value":497},"Anomaly records with ",{"type":39,"tag":76,"props":499,"children":501},{"className":500},[],[502],{"type":44,"value":503},"recordScore",{"type":44,"value":505},", ",{"type":39,"tag":76,"props":507,"children":509},{"className":508},[],[510],{"type":44,"value":511},"jobId",{"type":44,"value":505},{"type":39,"tag":76,"props":514,"children":516},{"className":515},[],[517],{"type":44,"value":518},"fieldName",{"type":44,"value":505},{"type":39,"tag":76,"props":521,"children":523},{"className":522},[],[524],{"type":44,"value":525},"functionName",{"type":44,"value":505},{"type":39,"tag":76,"props":528,"children":530},{"className":529},[],[531],{"type":44,"value":283},{"type":44,"value":505},{"type":39,"tag":76,"props":534,"children":536},{"className":535},[],[537],{"type":44,"value":538},"deviationPercent",{"type":44,"value":540},",\nand the actual vs typical values.",{"type":39,"tag":64,"props":542,"children":543},{},[544,546,552],{"type":44,"value":545},"A ",{"type":39,"tag":76,"props":547,"children":549},{"className":548},[],[550],{"type":44,"value":551},"jobsSummary",{"type":44,"value":553}," of counts per job.",{"type":39,"tag":64,"props":555,"children":556},{},[557,559,565],{"type":44,"value":558},"An ",{"type":39,"tag":76,"props":560,"children":562},{"className":561},[],[563],{"type":44,"value":564},"investigation_actions",{"type":44,"value":566}," list — pre-computed click-to-send follow-up prompts the view surfaces as buttons.",{"type":39,"tag":47,"props":568,"children":569},{},[570,572,578],{"type":44,"value":571},"Ignore ",{"type":39,"tag":76,"props":573,"children":575},{"className":574},[],[576],{"type":44,"value":577},"_setup_notice",{"type":44,"value":579}," if present — it's view-side chrome (welcome banner) that the UI handles. Don't\necho or summarize it in chat.",{"type":39,"tag":47,"props":581,"children":582},{},[583],{"type":44,"value":584},"The explainer view renders in one of two modes, picked automatically from the result shape:",{"type":39,"tag":60,"props":586,"children":587},{},[588,598],{"type":39,"tag":64,"props":589,"children":590},{},[591,596],{"type":39,"tag":68,"props":592,"children":593},{},[594],{"type":44,"value":595},"Overview mode",{"type":44,"value":597}," (many anomalies, cross-entity): severity counts, affected-entities list, by-ML-job breakdown.",{"type":39,"tag":64,"props":599,"children":600},{},[601,606],{"type":39,"tag":68,"props":602,"children":603},{},[604],{"type":44,"value":605},"Detail mode",{"type":44,"value":607}," (one anomaly, or filtered to a single entity): entity header, score \u002F actual \u002F typical \u002F\ndeviation cards, an actual-vs-typical comparison bar, and a time-series when available.",{"type":39,"tag":47,"props":609,"children":610},{},[611,616,618,623],{"type":39,"tag":68,"props":612,"children":613},{},[614],{"type":44,"value":615},"Use the view",{"type":44,"value":617}," — don't restate the JSON. Provide a narrative ",{"type":39,"tag":68,"props":619,"children":620},{},[621],{"type":44,"value":622},"below",{"type":44,"value":624}," it:",{"type":39,"tag":626,"props":627,"children":628},"ol",{},[629,646,656,673],{"type":39,"tag":64,"props":630,"children":631},{},[632,637,639,644],{"type":39,"tag":68,"props":633,"children":634},{},[635],{"type":44,"value":636},"Headline the worst offender",{"type":44,"value":638},": \"Top anomaly — ",{"type":39,"tag":76,"props":640,"children":642},{"className":641},[],[643],{"type":44,"value":300},{"type":44,"value":645}," memory working set anomalous, score 87\n(major), 340% above typical.\"",{"type":39,"tag":64,"props":647,"children":648},{},[649,654],{"type":39,"tag":68,"props":650,"children":651},{},[652],{"type":44,"value":653},"Group by entity",{"type":44,"value":655},": list the top 3-5 affected entities with one-line summaries (overview mode).",{"type":39,"tag":64,"props":657,"children":658},{},[659,664,666,671],{"type":39,"tag":68,"props":660,"children":661},{},[662],{"type":44,"value":663},"Respect the next-step buttons",{"type":44,"value":665},": the view shows ",{"type":39,"tag":76,"props":667,"children":669},{"className":668},[],[670],{"type":44,"value":564},{"type":44,"value":672}," as clickable prompts — call\nthem out in your reply (\"…or click Blast radius to see infra impact\") so the user knows they're there.",{"type":39,"tag":64,"props":674,"children":675},{},[676,681],{"type":39,"tag":68,"props":677,"children":678},{},[679],{"type":44,"value":680},"Flag gaps",{"type":44,"value":682},": if the user expected anomalies and none fired, say so — might mean jobs are behind or\nthresholds need tuning.",{"type":39,"tag":53,"props":684,"children":686},{"id":685},"key-principles",[687],{"type":44,"value":688},"Key principles",{"type":39,"tag":60,"props":690,"children":691},{},[692,702,712,722,753],{"type":39,"tag":64,"props":693,"children":694},{},[695,700],{"type":39,"tag":68,"props":696,"children":697},{},[698],{"type":44,"value":699},"Let the view do the visual work.",{"type":44,"value":701}," The explainer has a severity gauge and per-entity cards. Don't\nduplicate them in prose.",{"type":39,"tag":64,"props":703,"children":704},{},[705,710],{"type":39,"tag":68,"props":706,"children":707},{},[708],{"type":44,"value":709},"Anomaly score ≠ severity of the underlying issue.",{"type":44,"value":711}," A high score means \"unusual,\" not \"broken.\" Always\ncross-reference with what the user is actually seeing.",{"type":39,"tag":64,"props":713,"children":714},{},[715,720],{"type":39,"tag":68,"props":716,"children":717},{},[718],{"type":44,"value":719},"The ML baseline is what the jobs learned from the data's past.",{"type":44,"value":721}," Communicate anomalies as \"unusual\nvs typical behavior learned from prior N days,\" not as absolute verdicts.",{"type":39,"tag":64,"props":723,"children":724},{},[725,730,732,737,739,744,746,751],{"type":39,"tag":68,"props":726,"children":727},{},[728],{"type":44,"value":729},"Empty result is a signal, not a failure.",{"type":44,"value":731}," Treat an empty result as a definitive answer for the requested parameters — do ",{"type":39,"tag":68,"props":733,"children":734},{},[735],{"type":44,"value":736},"not",{"type":44,"value":738}," automatically re-run with ",{"type":39,"tag":76,"props":740,"children":742},{"className":741},[],[743],{"type":44,"value":334},{"type":44,"value":745}," lowered or ",{"type":39,"tag":76,"props":747,"children":749},{"className":748},[],[750],{"type":44,"value":240},{"type":44,"value":752}," widened. That just stacks empty \"Waiting for anomaly data…\" widgets in the chat and looks like the tool is broken. Instead, tell the user \"no anomalies above score X in the last Y\" and offer a single follow-up like \"Want me to broaden the search to min_score 25 \u002F 24h?\" — let them confirm before re-calling.",{"type":39,"tag":64,"props":754,"children":755},{},[756,768,770,775,777,783,785,790,792,798,800,805],{"type":39,"tag":68,"props":757,"children":758},{},[759,761,766],{"type":44,"value":760},"Pass ",{"type":39,"tag":76,"props":762,"children":764},{"className":763},[],[765],{"type":44,"value":283},{"type":44,"value":767}," as a plain value, not a composite.",{"type":44,"value":769}," The tool emits ",{"type":39,"tag":76,"props":771,"children":773},{"className":772},[],[774],{"type":44,"value":283},{"type":44,"value":776}," on each anomaly as ",{"type":39,"tag":76,"props":778,"children":780},{"className":779},[],[781],{"type":44,"value":782},"field1=value1; field2=value2",{"type":44,"value":784}," (the partition\u002Fby\u002Fover field info). Never paste that composite back in as the ",{"type":39,"tag":76,"props":786,"children":788},{"className":787},[],[789],{"type":44,"value":283},{"type":44,"value":791}," arg — pass just the value the user actually cares about (e.g. ",{"type":39,"tag":76,"props":793,"children":795},{"className":794},[],[796],{"type":44,"value":797},"kube-proxy-gke-...",{"type":44,"value":799}," or ",{"type":39,"tag":76,"props":801,"children":803},{"className":802},[],[804],{"type":44,"value":300},{"type":44,"value":806},"). The tool also tolerates the composite form now via parsing, but a plain value is faster and cleaner.",{"type":39,"tag":53,"props":808,"children":810},{"id":809},"investigation-discipline",[811],{"type":44,"value":812},"Investigation discipline",{"type":39,"tag":60,"props":814,"children":815},{},[816,826,844],{"type":39,"tag":64,"props":817,"children":818},{},[819,824],{"type":39,"tag":68,"props":820,"children":821},{},[822],{"type":44,"value":823},"One tool call per turn.",{"type":44,"value":825}," After this tool returns, narrate the headline finding — top entity, its score, what it means — before making another call. Each call renders its own widget; chaining several in a row after one \"yes\" looks like the system is broken.",{"type":39,"tag":64,"props":827,"children":828},{},[829,834,836,842],{"type":39,"tag":68,"props":830,"children":831},{},[832],{"type":44,"value":833},"Sequential offers, not OR.",{"type":44,"value":835}," Don't ask \"Want me to check anomalies for X ",{"type":39,"tag":837,"props":838,"children":839},"em",{},[840],{"type":44,"value":841},"or",{"type":44,"value":843}," Y?\" — phrase as \"I'll check X first; if it's quiet I'll move to Y.\" The user's \"yes\" then maps to one call, not both.",{"type":39,"tag":64,"props":845,"children":846},{},[847,852],{"type":39,"tag":68,"props":848,"children":849},{},[850],{"type":44,"value":851},"Don't auto-broaden.",{"type":44,"value":853}," If 0 anomalies for the requested params, say so and OFFER a wider search (\"Want me to widen to min_score 25 \u002F 24h?\"). Wait for confirmation before re-calling. Auto-retrying produces empty stacked widgets.",{"type":39,"tag":855,"props":856,"children":857},"style",{},[858],{"type":44,"value":859},"html .light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html.light .shiki span {color: var(--shiki-light);background: var(--shiki-light-bg);font-style: var(--shiki-light-font-style);font-weight: var(--shiki-light-font-weight);text-decoration: var(--shiki-light-text-decoration);}html .default .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .shiki span {color: var(--shiki-default);background: var(--shiki-default-bg);font-style: var(--shiki-default-font-style);font-weight: var(--shiki-default-font-weight);text-decoration: var(--shiki-default-text-decoration);}html .dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}html.dark .shiki span {color: var(--shiki-dark);background: var(--shiki-dark-bg);font-style: var(--shiki-dark-font-style);font-weight: var(--shiki-dark-font-weight);text-decoration: var(--shiki-dark-text-decoration);}",{"items":861,"total":935},[862,877,887,901,916,923],{"slug":863,"name":863,"fn":864,"description":865,"org":866,"tags":867,"stars":23,"repoUrl":24,"updatedAt":876},"apm-health-summary","summarize Elastic APM service health","Get a cluster-level rollup of service health from APM telemetry — the \"how's my environment right now?\" entry point for observability investigations. Use whenever the user asks about HEALTH, STATUS, or general wellbeing of an environment \u002F cluster \u002F namespace (\"how's my cluster\", \"status of the X env\", \"what's broken\", \"any issues\", \"show me the health of …\", \"give me a status report\", \"what should I look at\", \"things feel slow\"). This applies regardless of any time qualifier — \"show me the health of X over the past hour\" still routes here (with lookback=\"1h\"), NOT to observe. observe is for raw-metric queries; this tool is for the rollup. Gracefully degrades: layers in Kubernetes pod data and ML anomaly context when those backends are present, but still returns useful APM-only output if they aren't. Do not use for log-only or metrics-only customers — this tool requires Elastic APM.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[868,871,872,875],{"name":869,"slug":870,"type":15},"APM","apm",{"name":9,"slug":8,"type":15},{"name":873,"slug":874,"type":15},"Monitoring","monitoring",{"name":13,"slug":14,"type":15},"2026-07-12T07:49:24.405551",{"slug":174,"name":174,"fn":878,"description":879,"org":880,"tags":881,"stars":23,"repoUrl":24,"updatedAt":886},"map application topology from APM telemetry","Map the application topology from APM telemetry — which services call which, over what protocols, with what call volume and latency. Use when the user asks \"what calls X\", \"what depends on X\", \"show me the topology\", \"what are the upstream\u002Fdownstream services\", \"where does this service fit\", or is doing root-cause investigation and needs to trace how a problem propagates through the call graph. Also trigger for \"service map\", \"dependency graph\", \"blast radius of service X\", or \"who's the dependency of Y\". Requires Elastic APM — do not trigger for log-only or metrics-only customers.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[882,883,884,885],{"name":869,"slug":870,"type":15},{"name":9,"slug":8,"type":15},{"name":873,"slug":874,"type":15},{"name":13,"slug":14,"type":15},"2026-07-12T07:49:23.167442",{"slug":191,"name":191,"fn":888,"description":889,"org":890,"tags":891,"stars":23,"repoUrl":24,"updatedAt":900},"assess Kubernetes node failure impact","Assess the impact of a Kubernetes node going offline — which deployments lose all replicas (full outage), which lose partial capacity (degraded), which are unaffected, and whether the cluster has enough spare capacity to reschedule the lost pods. Use when the user asks \"what happens if node X goes down\", \"what's the blast radius of draining this node\", \"can I safely maintain node Y\", \"what's running on this node\", \"if I evict this node what breaks\", or is planning node maintenance, a cluster upgrade, or investigating an actual node failure. Requires Kubernetes (kubeletstats metrics) and Elastic APM for downstream service impact — do not trigger for non-K8s deployments.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[892,893,896,897],{"name":9,"slug":8,"type":15},{"name":894,"slug":895,"type":15},"Kubernetes","kubernetes",{"name":13,"slug":14,"type":15},{"name":898,"slug":899,"type":15},"Risk Assessment","risk-assessment","2026-07-12T07:49:25.645103",{"slug":902,"name":902,"fn":903,"description":904,"org":905,"tags":906,"stars":23,"repoUrl":24,"updatedAt":915},"manage-alerts","manage Kibana alerting rules","CRUD for Kibana alerting rules — create, list, get, or delete custom-threshold rules. Use when the user says \"alert me when\", \"create a rule for\", \"page me if\", \"set up an alert\", \"show me my rules\", \"what alerts do I have\", \"delete that alert\", \"remove the rule\". Backend-agnostic — works on any metric field in any index pattern (metrics-*, logs-*, traces-apm*, custom). For transient session-scoped monitoring use `observe` instead. Requires Kibana with the Alerting feature enabled — the tool is auto-disabled when no Kibana URL is configured.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[907,910,911,914],{"name":908,"slug":909,"type":15},"Alerting","alerting",{"name":9,"slug":8,"type":15},{"name":912,"slug":913,"type":15},"Kibana","kibana",{"name":873,"slug":874,"type":15},"2026-07-12T07:49:21.846108",{"slug":4,"name":4,"fn":5,"description":6,"org":917,"tags":918,"stars":23,"repoUrl":24,"updatedAt":25},{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[919,920,921,922],{"name":21,"slug":22,"type":15},{"name":9,"slug":8,"type":15},{"name":17,"slug":18,"type":15},{"name":13,"slug":14,"type":15},{"slug":155,"name":155,"fn":924,"description":925,"org":926,"tags":927,"stars":23,"repoUrl":24,"updatedAt":934},"monitor Elastic observability telemetry","The agent's Elastic-access primitive. Four modes: wait for an ML anomaly to fire, poll an ES|QL metric (live-sample or wait for a threshold), read a single-instance scalar value, or return a full ES|QL table. Use when the user says \"tell me when...\", \"let me know if...\", \"wait until X drops below Y\", \"watch for anything unusual\", \"monitor for the next N minutes\", \"poll until stable\", \"what is X right now\", \"list …\", \"which … are …\", or wants transient (session-scoped) monitoring or ad-hoc querying without creating a persistent Kibana rule. Also trigger for \"keep an eye on\" and post-remediation validation.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[928,929,932,933],{"name":9,"slug":8,"type":15},{"name":930,"slug":931,"type":15},"Metrics","metrics",{"name":873,"slug":874,"type":15},{"name":13,"slug":14,"type":15},"2026-07-12T07:49:28.114697",6,{"items":937,"total":1109},[938,957,974,989,1008,1020,1030,1045,1057,1072,1083,1096],{"slug":939,"name":939,"fn":940,"description":941,"org":942,"tags":943,"stars":954,"repoUrl":955,"updatedAt":956},"accessing-benchmark-results","retrieve and analyze Rally benchmark results","Retrieve Rally benchmark results from an external Elasticsearch metrics store. Use to list past races, get a single race's overall (per-task) results, chart a metric's trend across multiple runs, compare two races, or check whether a run converged — e.g. \"show me recent geonames races\", \"what's the service_time trend for nyc_taxis over the last 30 days?\", \"compare these two race-ids\". Applies when datastore.type = elasticsearch is set in ~\u002F.rally\u002Frally.ini.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[944,947,950,951],{"name":945,"slug":946,"type":15},"Analytics","analytics",{"name":948,"slug":949,"type":15},"Data Analysis","data-analysis",{"name":9,"slug":8,"type":15},{"name":952,"slug":953,"type":15},"Performance","performance",2027,"https:\u002F\u002Fgithub.com\u002Felastic\u002Frally","2026-07-12T07:46:38.54144",{"slug":958,"name":958,"fn":959,"description":960,"org":961,"tags":962,"stars":954,"repoUrl":955,"updatedAt":973},"developing-rally","develop and debug Rally source code","Work on Rally's own codebase, not running benchmarks with it. Use when setting up the dev environment, running Rally's tests or linters, navigating its source, debugging Rally's own code, or making changes to Rally itself.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[963,966,967,970],{"name":964,"slug":965,"type":15},"Debugging","debugging",{"name":9,"slug":8,"type":15},{"name":968,"slug":969,"type":15},"Engineering","engineering",{"name":971,"slug":972,"type":15},"Local Development","local-development","2026-07-12T07:46:35.976807",{"slug":975,"name":975,"fn":976,"description":977,"org":978,"tags":979,"stars":954,"repoUrl":955,"updatedAt":988},"running-benchmarks","run Rally benchmarks against Elasticsearch","Run Rally benchmarks (races) against Elasticsearch — an existing\u002Fexternal cluster or a Rally-provisioned distribution — and read the summary report. Use when running a race (any pipeline, track, challenge, target-hosts, or auth) or when interpreting throughput, latency, and service_time results.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[980,981,984,985],{"name":9,"slug":8,"type":15},{"name":982,"slug":983,"type":15},"Elasticsearch","elasticsearch",{"name":952,"slug":953,"type":15},{"name":986,"slug":987,"type":15},"Testing","testing","2026-07-12T07:46:37.277964",{"slug":990,"name":990,"fn":991,"description":992,"org":993,"tags":994,"stars":1005,"repoUrl":1006,"updatedAt":1007},"cloud-access-management","manage Elastic Cloud organization access","Manage Elastic Cloud organization access: invite users, assign roles to Serverless projects, and create or revoke Cloud API keys. Use when granting, modifying, or auditing user access.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[995,998,999,1002],{"name":996,"slug":997,"type":15},"Cloud","cloud",{"name":9,"slug":8,"type":15},{"name":1000,"slug":1001,"type":15},"Operations","operations",{"name":1003,"slug":1004,"type":15},"Permissions","permissions",531,"https:\u002F\u002Fgithub.com\u002Felastic\u002Fagent-skills","2026-07-12T07:46:44.946285",{"slug":1009,"name":1009,"fn":1010,"description":1011,"org":1012,"tags":1013,"stars":1005,"repoUrl":1006,"updatedAt":1019},"cloud-create-project","create Elastic Cloud Serverless projects","Creates Elastic Cloud Serverless projects (Elasticsearch, Observability, or Security) via the REST API, saves credentials to file, and bootstraps a scoped Elasticsearch API key. Use when creating a new serverless project, provisioning a search or observability environment, or spinning up a new Elastic Cloud project.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1014,1015,1018],{"name":996,"slug":997,"type":15},{"name":1016,"slug":1017,"type":15},"Deployment","deployment",{"name":982,"slug":983,"type":15},"2026-07-12T07:46:42.353362",{"slug":1021,"name":1021,"fn":1022,"description":1023,"org":1024,"tags":1025,"stars":1005,"repoUrl":1006,"updatedAt":1029},"cloud-manage-project","manage Elastic Cloud Serverless projects","Manages existing Elastic Cloud Serverless projects: list, get, update, delete, reset credentials, resume, and load saved credentials. Connects to existing projects by resolving endpoints and acquiring scoped Elasticsearch API keys. Use when performing day-2 operations on serverless projects, connecting to an existing project, loading or resetting project credentials, or looking up project details.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1026,1027,1028],{"name":996,"slug":997,"type":15},{"name":982,"slug":983,"type":15},{"name":1000,"slug":1001,"type":15},"2026-07-12T07:46:41.097412",{"slug":1031,"name":1031,"fn":1032,"description":1033,"org":1034,"tags":1035,"stars":1005,"repoUrl":1006,"updatedAt":1044},"cloud-network-security","manage Elastic Cloud network security","Manage Serverless network security (traffic filters): create, update, and delete IP filters and AWS PrivateLink VPC filters. Use when restricting network access or configuring private connectivity.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1036,1037,1038,1041],{"name":996,"slug":997,"type":15},{"name":982,"slug":983,"type":15},{"name":1039,"slug":1040,"type":15},"Networking","networking",{"name":1042,"slug":1043,"type":15},"Security","security","2026-07-12T07:46:43.675992",{"slug":1046,"name":1046,"fn":1047,"description":1048,"org":1049,"tags":1050,"stars":1005,"repoUrl":1006,"updatedAt":1056},"cloud-setup","configure Elastic Cloud authentication","Configures Elastic Cloud authentication and environment defaults. Use when setting up EC_API_KEY, configuring Cloud API access, or when another cloud skill requires credentials.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1051,1054,1055],{"name":1052,"slug":1053,"type":15},"Authentication","authentication",{"name":996,"slug":997,"type":15},{"name":982,"slug":983,"type":15},"2026-07-12T07:46:39.783105",{"slug":1058,"name":1058,"fn":1059,"description":1060,"org":1061,"tags":1062,"stars":1005,"repoUrl":1006,"updatedAt":1071},"elasticsearch-audit","configure Elasticsearch security audit logs","Enable, configure, and query Elasticsearch security audit logs. Use when the task involves audit logging setup, event filtering, or investigating security incidents like failed logins.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1063,1066,1067,1070],{"name":1064,"slug":1065,"type":15},"Audit","audit",{"name":982,"slug":983,"type":15},{"name":1068,"slug":1069,"type":15},"Logs","logs",{"name":1042,"slug":1043,"type":15},"2026-07-12T07:47:35.092599",{"slug":1073,"name":1073,"fn":1074,"description":1075,"org":1076,"tags":1077,"stars":1005,"repoUrl":1006,"updatedAt":1082},"elasticsearch-authn","configure Elasticsearch authentication realms","Authenticate to Elasticsearch using native, file-based, LDAP\u002FAD, SAML, OIDC, Kerberos, JWT, or certificate realms. Use when connecting with credentials, choosing a realm, or managing API keys. Assumes the target realms are already configured.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1078,1079,1080,1081],{"name":1052,"slug":1053,"type":15},{"name":9,"slug":8,"type":15},{"name":982,"slug":983,"type":15},{"name":1042,"slug":1043,"type":15},"2026-07-12T07:47:41.474547",{"slug":1084,"name":1084,"fn":1085,"description":1086,"org":1087,"tags":1088,"stars":1005,"repoUrl":1006,"updatedAt":1095},"elasticsearch-authz","manage Elasticsearch RBAC and security roles","Manage Elasticsearch RBAC: native users, roles, role mappings, document- and field-level security. Use when creating users or roles, assigning privileges, or mapping external realms like LDAP\u002FSAML.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1089,1090,1091,1094],{"name":9,"slug":8,"type":15},{"name":982,"slug":983,"type":15},{"name":1092,"slug":1093,"type":15},"RBAC","rbac",{"name":1042,"slug":1043,"type":15},"2026-07-12T07:47:36.394177",{"slug":1097,"name":1097,"fn":1098,"description":1099,"org":1100,"tags":1101,"stars":1005,"repoUrl":1006,"updatedAt":1108},"elasticsearch-esql","query Elasticsearch data with ES|QL","Execute ES|QL (Elasticsearch Query Language) queries, use when the user wants to query Elasticsearch data, analyze logs, aggregate metrics, explore data, or create charts and dashboards from ES|QL results.\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":8},[1102,1103,1104,1105],{"name":945,"slug":946,"type":15},{"name":948,"slug":949,"type":15},{"name":982,"slug":983,"type":15},{"name":1106,"slug":1107,"type":15},"SQL","sql","2026-07-12T07:47:40.249533",86]