[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill-braintrust-braintrust-elicit-eval-criteria":3,"mdc-pb8k8q-key":38,"related-org-braintrust-braintrust-elicit-eval-criteria":336,"related-repo-braintrust-braintrust-elicit-eval-criteria":509},{"slug":4,"name":4,"fn":5,"description":6,"org":7,"tags":12,"stars":27,"repoUrl":28,"updatedAt":29,"license":30,"forks":31,"topics":32,"repo":33,"sourceUrl":36,"mdContent":37},"braintrust-elicit-eval-criteria","extract and define evaluation criteria","Extract evaluation criteria out of domain experts and real user desires, and capture them as reusable evaluation assets before any labeling or scoring begins — construct facets, anchored exemplars, adversarial traps, scoring guidance, audit rules, and the signals that reveal what users actually want. Use when nobody can say what \"good\" means, when a rubric does not exist yet, when expert knowledge lives only in reviewers' heads, or when validating that an eval reflects user desires rather than team assumptions. Do not use to run the labeling workflow or to compare a scorer against finished labels.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},"braintrust","Braintrust","https:\u002F\u002Fpexgzepcugksgbtrxkhf.supabase.co\u002Fstorage\u002Fv1\u002Fobject\u002Fpublic\u002Forg-logos\u002Fbraintrust.png","braintrustdata",[13,17,20,23,26],{"name":14,"slug":15,"type":16},"LLM","llm","tag",{"name":18,"slug":19,"type":16},"Ideation","ideation",{"name":21,"slug":22,"type":16},"Evals","evals",{"name":24,"slug":25,"type":16},"Strategy","strategy",{"name":9,"slug":8,"type":16},7,"https:\u002F\u002Fgithub.com\u002Fbraintrustdata\u002Feval-library","2026-08-20T03:53:36.164477",null,0,[],{"repoUrl":28,"stars":27,"forks":31,"topics":34,"description":35},[],"Braintrust eval skills library","https:\u002F\u002Fgithub.com\u002Fbraintrustdata\u002Feval-library\u002Ftree\u002FHEAD\u002Fskills\u002Fbraintrust-elicit-eval-criteria","---\nname: braintrust-elicit-eval-criteria\ndescription: >-\n  Extract evaluation criteria out of domain experts and real user desires, and capture them as\n  reusable evaluation assets before any labeling or scoring begins — construct facets, anchored\n  exemplars, adversarial traps, scoring guidance, audit rules, and the signals that reveal what\n  users actually want. Use when nobody can say what \"good\" means, when a rubric does not exist\n  yet, when expert knowledge lives only in reviewers' heads, or when validating that an eval\n  reflects user desires rather than team assumptions. Do not use to run the labeling workflow or\n  to compare a scorer against finished labels.\n---\n\n# Elicit criteria from experts and users\n\nContract: `references\u002Finteraction-contract.md`. Calibration, templates, provenance: `references\u002Felicitation-guide.md`.\n\n## Trigger\n\n- No rubric exists, and review is happening on instinct.\n- Experts disagree about what \"good\" means, or agree without being able to say why.\n- An eval built entirely from team assumptions about what users want.\n\n## Do\n\n1. Gather both sources. **Expert knowledge**: what a qualified reviewer knows that the rubric\n   does not say. **User desire**: what users actually want, evidenced rather than assumed.\n2. Elicit from cases, not abstractions. Put a real trace in front of the expert, ask for a\n   judgment, then ask *why* — the reason is the criterion. Repeat until reasons stop being new.\n3. Capture each facet as an **anchored exemplar pair**: one output earning the top score, one\n   that just misses, and the distinguishing reason. A pair beats a paragraph of description.\n4. Harvest user desire from evidence that already exists — escalations, reopened tickets,\n   abandoned sessions, thumbs-down, support themes. Where none exists, say so and label the\n   criteria `Assumed`.\n5. Capture what experts know about **failure**: the traps, the plausible-looking wrong answer,\n   what juniors get wrong. Highest-value output, and the part rubrics usually omit.\n\n## Avoid\n\n- Do not ask experts to write a rubric from a blank page; elicit from cases.\n- Do not substitute team intuition for user desire, or present it as evidence when you do.\n- Do not run the labeling workflow or implement scoring here; this stage produces the criteria\n  those steps consume.\n- Do not resolve genuine expert disagreement by averaging it — a contested criterion is a\n  finding about the construct.\n\n## Check\n\n- Every facet has an anchored exemplar pair with the distinguishing reason.\n- Failure knowledge captured explicitly, not just success criteria.\n- Each criterion labeled `Confirmed` (evidenced) or `Assumed` (team belief).\n- User-desire signals named with their source, or their absence stated.\n\n## Risk\n\n- Experts articulate what they can defend, not always what they use; the reason given may not be\n  the reason applied, which is why exemplars beat prose.\n- A single expert encodes one house style as a universal standard.\n- Available evidence of desire is biased toward users who complain; silent dissatisfaction leaves\n  no trace.\n- Criteria elicited once ossify — they need the dataset's refresh cadence.\n\n## Braintrust\n\nRun the elicitation session itself as **human review** over real traces with **per-criterion\nscoring and free-text notes** — the notes are the output, because they carry the *reason*.\nPromote reviewed traces into a **versioned golden dataset** so exemplars travel with the\ncriteria instead of living in a doc. Each facet then becomes its own scorer with a consistent\nname; anchored pairs become the few-shot content of a rubric scorer. Flag traps in `metadata`\n(`trap: true`) so they can be sliced out of headline numbers and reported separately. Treat a\ncriteria revision like a scorer change: bump the version, and do not compare results across the\nboundary.\n",{"data":39,"body":40},{"name":4,"description":6},{"type":41,"children":42},"root",[43,52,75,82,102,108,181,187,210,216,254,260,283,287],{"type":44,"tag":45,"props":46,"children":48},"element","h1",{"id":47},"elicit-criteria-from-experts-and-users",[49],{"type":50,"value":51},"text","Elicit criteria from experts and users",{"type":44,"tag":53,"props":54,"children":55},"p",{},[56,58,65,67,73],{"type":50,"value":57},"Contract: ",{"type":44,"tag":59,"props":60,"children":62},"code",{"className":61},[],[63],{"type":50,"value":64},"references\u002Finteraction-contract.md",{"type":50,"value":66},". Calibration, templates, provenance: ",{"type":44,"tag":59,"props":68,"children":70},{"className":69},[],[71],{"type":50,"value":72},"references\u002Felicitation-guide.md",{"type":50,"value":74},".",{"type":44,"tag":76,"props":77,"children":79},"h2",{"id":78},"trigger",[80],{"type":50,"value":81},"Trigger",{"type":44,"tag":83,"props":84,"children":85},"ul",{},[86,92,97],{"type":44,"tag":87,"props":88,"children":89},"li",{},[90],{"type":50,"value":91},"No rubric exists, and review is happening on instinct.",{"type":44,"tag":87,"props":93,"children":94},{},[95],{"type":50,"value":96},"Experts disagree about what \"good\" means, or agree without being able to say why.",{"type":44,"tag":87,"props":98,"children":99},{},[100],{"type":50,"value":101},"An eval built entirely from team assumptions about what users want.",{"type":44,"tag":76,"props":103,"children":105},{"id":104},"do",[106],{"type":50,"value":107},"Do",{"type":44,"tag":109,"props":110,"children":111},"ol",{},[112,132,145,157,169],{"type":44,"tag":87,"props":113,"children":114},{},[115,117,123,125,130],{"type":50,"value":116},"Gather both sources. ",{"type":44,"tag":118,"props":119,"children":120},"strong",{},[121],{"type":50,"value":122},"Expert knowledge",{"type":50,"value":124},": what a qualified reviewer knows that the rubric\ndoes not say. ",{"type":44,"tag":118,"props":126,"children":127},{},[128],{"type":50,"value":129},"User desire",{"type":50,"value":131},": what users actually want, evidenced rather than assumed.",{"type":44,"tag":87,"props":133,"children":134},{},[135,137,143],{"type":50,"value":136},"Elicit from cases, not abstractions. Put a real trace in front of the expert, ask for a\njudgment, then ask ",{"type":44,"tag":138,"props":139,"children":140},"em",{},[141],{"type":50,"value":142},"why",{"type":50,"value":144}," — the reason is the criterion. Repeat until reasons stop being new.",{"type":44,"tag":87,"props":146,"children":147},{},[148,150,155],{"type":50,"value":149},"Capture each facet as an ",{"type":44,"tag":118,"props":151,"children":152},{},[153],{"type":50,"value":154},"anchored exemplar pair",{"type":50,"value":156},": one output earning the top score, one\nthat just misses, and the distinguishing reason. A pair beats a paragraph of description.",{"type":44,"tag":87,"props":158,"children":159},{},[160,162,168],{"type":50,"value":161},"Harvest user desire from evidence that already exists — escalations, reopened tickets,\nabandoned sessions, thumbs-down, support themes. Where none exists, say so and label the\ncriteria ",{"type":44,"tag":59,"props":163,"children":165},{"className":164},[],[166],{"type":50,"value":167},"Assumed",{"type":50,"value":74},{"type":44,"tag":87,"props":170,"children":171},{},[172,174,179],{"type":50,"value":173},"Capture what experts know about ",{"type":44,"tag":118,"props":175,"children":176},{},[177],{"type":50,"value":178},"failure",{"type":50,"value":180},": the traps, the plausible-looking wrong answer,\nwhat juniors get wrong. Highest-value output, and the part rubrics usually omit.",{"type":44,"tag":76,"props":182,"children":184},{"id":183},"avoid",[185],{"type":50,"value":186},"Avoid",{"type":44,"tag":83,"props":188,"children":189},{},[190,195,200,205],{"type":44,"tag":87,"props":191,"children":192},{},[193],{"type":50,"value":194},"Do not ask experts to write a rubric from a blank page; elicit from cases.",{"type":44,"tag":87,"props":196,"children":197},{},[198],{"type":50,"value":199},"Do not substitute team intuition for user desire, or present it as evidence when you do.",{"type":44,"tag":87,"props":201,"children":202},{},[203],{"type":50,"value":204},"Do not run the labeling workflow or implement scoring here; this stage produces the criteria\nthose steps consume.",{"type":44,"tag":87,"props":206,"children":207},{},[208],{"type":50,"value":209},"Do not resolve genuine expert disagreement by averaging it — a contested criterion is a\nfinding about the construct.",{"type":44,"tag":76,"props":211,"children":213},{"id":212},"check",[214],{"type":50,"value":215},"Check",{"type":44,"tag":83,"props":217,"children":218},{},[219,224,229,249],{"type":44,"tag":87,"props":220,"children":221},{},[222],{"type":50,"value":223},"Every facet has an anchored exemplar pair with the distinguishing reason.",{"type":44,"tag":87,"props":225,"children":226},{},[227],{"type":50,"value":228},"Failure knowledge captured explicitly, not just success criteria.",{"type":44,"tag":87,"props":230,"children":231},{},[232,234,240,242,247],{"type":50,"value":233},"Each criterion labeled ",{"type":44,"tag":59,"props":235,"children":237},{"className":236},[],[238],{"type":50,"value":239},"Confirmed",{"type":50,"value":241}," (evidenced) or ",{"type":44,"tag":59,"props":243,"children":245},{"className":244},[],[246],{"type":50,"value":167},{"type":50,"value":248}," (team belief).",{"type":44,"tag":87,"props":250,"children":251},{},[252],{"type":50,"value":253},"User-desire signals named with their source, or their absence stated.",{"type":44,"tag":76,"props":255,"children":257},{"id":256},"risk",[258],{"type":50,"value":259},"Risk",{"type":44,"tag":83,"props":261,"children":262},{},[263,268,273,278],{"type":44,"tag":87,"props":264,"children":265},{},[266],{"type":50,"value":267},"Experts articulate what they can defend, not always what they use; the reason given may not be\nthe reason applied, which is why exemplars beat prose.",{"type":44,"tag":87,"props":269,"children":270},{},[271],{"type":50,"value":272},"A single expert encodes one house style as a universal standard.",{"type":44,"tag":87,"props":274,"children":275},{},[276],{"type":50,"value":277},"Available evidence of desire is biased toward users who complain; silent dissatisfaction leaves\nno trace.",{"type":44,"tag":87,"props":279,"children":280},{},[281],{"type":50,"value":282},"Criteria elicited once ossify — they need the dataset's refresh cadence.",{"type":44,"tag":76,"props":284,"children":285},{"id":8},[286],{"type":50,"value":9},{"type":44,"tag":53,"props":288,"children":289},{},[290,292,297,299,304,306,311,313,318,320,326,328,334],{"type":50,"value":291},"Run the elicitation session itself as ",{"type":44,"tag":118,"props":293,"children":294},{},[295],{"type":50,"value":296},"human review",{"type":50,"value":298}," over real traces with ",{"type":44,"tag":118,"props":300,"children":301},{},[302],{"type":50,"value":303},"per-criterion\nscoring and free-text notes",{"type":50,"value":305}," — the notes are the output, because they carry the ",{"type":44,"tag":138,"props":307,"children":308},{},[309],{"type":50,"value":310},"reason",{"type":50,"value":312},".\nPromote reviewed traces into a ",{"type":44,"tag":118,"props":314,"children":315},{},[316],{"type":50,"value":317},"versioned golden dataset",{"type":50,"value":319}," so exemplars travel with the\ncriteria instead of living in a doc. Each facet then becomes its own scorer with a consistent\nname; anchored pairs become the few-shot content of a rubric scorer. Flag traps in ",{"type":44,"tag":59,"props":321,"children":323},{"className":322},[],[324],{"type":50,"value":325},"metadata",{"type":50,"value":327},"\n(",{"type":44,"tag":59,"props":329,"children":331},{"className":330},[],[332],{"type":50,"value":333},"trap: true",{"type":50,"value":335},") so they can be sliced out of headline numbers and reported separately. Treat a\ncriteria revision like a scorer change: bump the version, and do not compare results across the\nboundary.",{"items":337,"total":508},[338,354,370,384,400,414,428,440,454,470,482,494],{"slug":339,"name":339,"fn":340,"description":341,"org":342,"tags":343,"stars":351,"repoUrl":352,"updatedAt":353},"troubleshoot-braintrust-mcp","configure and troubleshoot Braintrust MCP servers","This plugin auto-configures a \"braintrust\" MCP server. If you can't see it or reach it, activate this skill\n",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[344,345,348],{"name":9,"slug":8,"type":16},{"name":346,"slug":347,"type":16},"Debugging","debugging",{"name":349,"slug":350,"type":16},"MCP","mcp",18,"https:\u002F\u002Fgithub.com\u002Fbraintrustdata\u002Fbraintrust-claude-plugin","2026-07-12T08:36:13.889274",{"slug":355,"name":355,"fn":356,"description":357,"org":358,"tags":359,"stars":27,"repoUrl":28,"updatedAt":369},"braintrust-analyze-eval-experiment","analyze LLM and agent eval experiments","Analyze completed LLM or agent eval experiments using uncertainty-aware and decision-relevant methods. Use to audit run completeness and pairing, calculate confidence intervals, run paired comparisons, report wins, losses, and ties, incorporate run-to-run variance, handle multiple comparisons, inspect subgroup performance, and test fragility to favorable slices. Use when results already exist and someone asks what they mean, whether a difference is real, or which model won. Do not use to design an experiment that has not yet collected results.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[360,363,364,365,366],{"name":361,"slug":362,"type":16},"Analysis","analysis",{"name":9,"slug":8,"type":16},{"name":21,"slug":22,"type":16},{"name":14,"slug":15,"type":16},{"name":367,"slug":368,"type":16},"Statistics","statistics","2026-08-20T03:53:01.13806",{"slug":371,"name":371,"fn":372,"description":373,"org":374,"tags":375,"stars":27,"repoUrl":28,"updatedAt":383},"braintrust-attribute-multi-variable-change","attribute performance changes to multiple variables","Attribute an observed change when several things moved at once — model plus prompt plus tools, a provider migration, a framework upgrade, or a vendor swap that bundles serving stack with model. Use when asked which part of a change caused the result, when a comparison's arms differ in more than one way, when a treatment has no uniform implementation across vendors, or when a serving-stack difference is confounded with a model difference. Do not use for a clean single-variable comparison, or to design an experiment that has not yet run.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[376,377,378,379,380],{"name":361,"slug":362,"type":16},{"name":9,"slug":8,"type":16},{"name":21,"slug":22,"type":16},{"name":14,"slug":15,"type":16},{"name":381,"slug":382,"type":16},"Performance","performance","2026-08-20T03:53:40.036077",{"slug":385,"name":385,"fn":386,"description":387,"org":388,"tags":389,"stars":27,"repoUrl":28,"updatedAt":399},"braintrust-build-eval-dataset","create and manage LLM eval datasets","Create, edit, audit, or compare eval datasets for LLM applications and agents, including target-population definition, case sourcing from production traces, stratified sampling, label provenance and label audits, expected values as constraints for open-ended tasks, dev\u002Ftest splits, contamination and leakage controls, headroom checks, refresh policy, and datasheets. Use when working on the content or lifecycle of an eval dataset. Do not use for sample-size or power calculations, scorer implementation, or open-ended adversarial failure discovery.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[390,393,394,397,398],{"name":391,"slug":392,"type":16},"Agents","agents",{"name":9,"slug":8,"type":16},{"name":395,"slug":396,"type":16},"Datasets","datasets",{"name":21,"slug":22,"type":16},{"name":14,"slug":15,"type":16},"2026-08-20T03:53:33.304815",{"slug":401,"name":401,"fn":402,"description":403,"org":404,"tags":405,"stars":27,"repoUrl":28,"updatedAt":413},"braintrust-define-eval-objective","define LLM evaluation objectives","Create, edit, or audit an eval objective by working backward from a product decision to the target outcome, construct, population, intended claim, and verification-versus-validation questions. Use when a team is unsure what an eval should establish, asks \"what are we actually trying to measure,\" \"is this eval measuring the right thing,\" \"does this benchmark support our claim,\" or needs to turn a product goal into an eval objective and state which claims are out of scope. Do not use to select detailed metrics, design datasets, or implement scorers.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[406,407,408,409,412],{"name":9,"slug":8,"type":16},{"name":21,"slug":22,"type":16},{"name":14,"slug":15,"type":16},{"name":410,"slug":411,"type":16},"Product Management","product-management",{"name":24,"slug":25,"type":16},"2026-08-20T03:53:00.07097",{"slug":415,"name":415,"fn":416,"description":417,"org":418,"tags":419,"stars":27,"repoUrl":28,"updatedAt":427},"braintrust-define-eval-release-gate","configure release gates for LLM applications","Create, edit, audit, or apply release gates for LLM applications and agents. Use to combine minimum meaningful improvement, statistical significance, regression rate, subgroup consistency, worst-run stability, all-attempts reliability, safety upper bounds, latency, and cost into an explicit ship-or-hold policy, to turn metrics into a CI gate, or to explain why a candidate failed a gate and what evidence would justify reconsideration. Do not use for general result analysis without a deployment decision.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[420,421,422,425,426],{"name":391,"slug":392,"type":16},{"name":9,"slug":8,"type":16},{"name":423,"slug":424,"type":16},"CI\u002FCD","ci-cd",{"name":21,"slug":22,"type":16},{"name":14,"slug":15,"type":16},"2026-08-20T03:53:15.686158",{"slug":429,"name":429,"fn":430,"description":431,"org":432,"tags":433,"stars":27,"repoUrl":28,"updatedAt":439},"braintrust-deploy-evaluator","deploy evaluators to Braintrust","Take a validated scorer or classifier from definition to running instrument in Braintrust — scope selection, inline testing before saving, saving as an evaluator, attaching an online-scoring rule, activating it for new traffic, and backfilling history with a rewind. Use when a scorer needs to actually run against production logs, when an online-scoring rule needs to be created or changed, or when historical traces need scoring. Do not use to decide what the scorer should measure, to write its rubric, or to establish that it agrees with human judgment.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[434,435,438],{"name":9,"slug":8,"type":16},{"name":436,"slug":437,"type":16},"Deployment","deployment",{"name":21,"slug":22,"type":16},"2026-08-20T03:53:32.558937",{"slug":441,"name":441,"fn":442,"description":443,"org":444,"tags":445,"stars":27,"repoUrl":28,"updatedAt":453},"braintrust-design-eval-experiment","design controlled LLM eval experiments","Design or audit controlled eval experiments for model, prompt, retrieval, tool, guardrail, or agent-architecture changes. Use before data collection to state directional and minimum-effect hypotheses, name independent, dependent, and control variables including the serving environment and tool surface, choose paired designs, set repetitions and allocation, distinguish exploratory from confirmatory comparisons, and pre-specify stopping, exclusion, multiplicity, and analysis rules. Do not use primarily to analyze results already collected.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[446,447,448,451,452],{"name":9,"slug":8,"type":16},{"name":21,"slug":22,"type":16},{"name":449,"slug":450,"type":16},"Experiments","experiments",{"name":14,"slug":15,"type":16},{"name":24,"slug":25,"type":16},"2026-08-20T03:53:36.534554",{"slug":455,"name":455,"fn":456,"description":457,"org":458,"tags":459,"stars":27,"repoUrl":28,"updatedAt":469},"braintrust-design-eval-instrumentation","design trace and evaluation dataset schemas","Design the trace and eval-dataset schema for an LLM app or agent, and wire the system to emit it. Use when deciding what to log, designing a trace schema, setting up tracing or observability before evals, or when failures cannot be debugged or sliced from existing traces — covering inputs, outputs, spans for tool and LLM calls, state changes, metadata, resolved configuration, serving path, tool manifest, per-item status, attachments, and subgroup variables. Do not use to decide what the evidence should mean, or to write scorers.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[460,461,462,463,466],{"name":9,"slug":8,"type":16},{"name":395,"slug":396,"type":16},{"name":21,"slug":22,"type":16},{"name":464,"slug":465,"type":16},"Observability","observability",{"name":467,"slug":468,"type":16},"Tracing","tracing","2026-08-20T03:53:37.274703",{"slug":471,"name":471,"fn":472,"description":473,"org":474,"tags":475,"stars":27,"repoUrl":28,"updatedAt":481},"braintrust-design-eval-metric-bundle","create multi-objective evaluation metric bundles","Create, edit, audit, or compare a multi-objective eval metric bundle covering product quality, safety, reliability, latency, and cost. Use to choose metrics for an eval, define a goodness bundle, distinguish optimization metrics from non-regression guardrails, expose tradeoffs, audit a KPI or single composite score for Goodhart and metric-gaming risk, or answer \"what should improve and what must not regress.\" Do not use to design trace schemas, build datasets, or implement scoring methods.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[476,477,478,479,480],{"name":391,"slug":392,"type":16},{"name":9,"slug":8,"type":16},{"name":21,"slug":22,"type":16},{"name":14,"slug":15,"type":16},{"name":381,"slug":382,"type":16},"2026-08-20T03:53:16.078609",{"slug":483,"name":483,"fn":484,"description":485,"org":486,"tags":487,"stars":27,"repoUrl":28,"updatedAt":493},"braintrust-design-human-eval-review","design human evaluation and review workflows","Design or audit human evaluation workflows and golden datasets for LLM applications and agents. Use to set up expert review, select review cases, write reviewer instructions, assign raters, capture rationales and confidence, measure inter-rater agreement with kappa or alpha, adjudicate disagreements, and preserve reviewed examples with provenance as a versioned reference set. Do not use to elicit the criteria or rubric in the first place, to validate a scorer once reference labels exist, or to implement the scorer.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[488,489,490,491,492],{"name":391,"slug":392,"type":16},{"name":9,"slug":8,"type":16},{"name":395,"slug":396,"type":16},{"name":21,"slug":22,"type":16},{"name":14,"slug":15,"type":16},"2026-08-20T03:53:36.900129",{"slug":495,"name":495,"fn":496,"description":497,"org":498,"tags":499,"stars":27,"repoUrl":28,"updatedAt":507},"braintrust-discover-agent-failures","identify and classify agent failure modes","Search open-endedly for unanticipated agent failure modes and convert them into a named taxonomy and durable regression items. Use for requests to find out what goes wrong, surface unknown or silent failures, do error analysis over traces, cluster and triage production failures, or build a failure taxonomy — where the goal is discovering modes nobody thought to test rather than measuring a predefined criterion. Produces datasets and taxonomies, not headline scores. Do not use for adversarial attacks against a threat model, or to measure a known criterion.",{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[500,501,502,503,504],{"name":391,"slug":392,"type":16},{"name":9,"slug":8,"type":16},{"name":346,"slug":347,"type":16},{"name":21,"slug":22,"type":16},{"name":505,"slug":506,"type":16},"Triage","triage","2026-08-20T03:53:00.423941",27,{"items":510,"total":565},[511,519,527,535,543,551,557],{"slug":355,"name":355,"fn":356,"description":357,"org":512,"tags":513,"stars":27,"repoUrl":28,"updatedAt":369},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[514,515,516,517,518],{"name":361,"slug":362,"type":16},{"name":9,"slug":8,"type":16},{"name":21,"slug":22,"type":16},{"name":14,"slug":15,"type":16},{"name":367,"slug":368,"type":16},{"slug":371,"name":371,"fn":372,"description":373,"org":520,"tags":521,"stars":27,"repoUrl":28,"updatedAt":383},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[522,523,524,525,526],{"name":361,"slug":362,"type":16},{"name":9,"slug":8,"type":16},{"name":21,"slug":22,"type":16},{"name":14,"slug":15,"type":16},{"name":381,"slug":382,"type":16},{"slug":385,"name":385,"fn":386,"description":387,"org":528,"tags":529,"stars":27,"repoUrl":28,"updatedAt":399},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[530,531,532,533,534],{"name":391,"slug":392,"type":16},{"name":9,"slug":8,"type":16},{"name":395,"slug":396,"type":16},{"name":21,"slug":22,"type":16},{"name":14,"slug":15,"type":16},{"slug":401,"name":401,"fn":402,"description":403,"org":536,"tags":537,"stars":27,"repoUrl":28,"updatedAt":413},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[538,539,540,541,542],{"name":9,"slug":8,"type":16},{"name":21,"slug":22,"type":16},{"name":14,"slug":15,"type":16},{"name":410,"slug":411,"type":16},{"name":24,"slug":25,"type":16},{"slug":415,"name":415,"fn":416,"description":417,"org":544,"tags":545,"stars":27,"repoUrl":28,"updatedAt":427},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[546,547,548,549,550],{"name":391,"slug":392,"type":16},{"name":9,"slug":8,"type":16},{"name":423,"slug":424,"type":16},{"name":21,"slug":22,"type":16},{"name":14,"slug":15,"type":16},{"slug":429,"name":429,"fn":430,"description":431,"org":552,"tags":553,"stars":27,"repoUrl":28,"updatedAt":439},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[554,555,556],{"name":9,"slug":8,"type":16},{"name":436,"slug":437,"type":16},{"name":21,"slug":22,"type":16},{"slug":441,"name":441,"fn":442,"description":443,"org":558,"tags":559,"stars":27,"repoUrl":28,"updatedAt":453},{"slug":8,"name":9,"logoUrl":10,"githubOrg":11},[560,561,562,563,564],{"name":9,"slug":8,"type":16},{"name":21,"slug":22,"type":16},{"name":449,"slug":450,"type":16},{"name":14,"slug":15,"type":16},{"name":24,"slug":25,"type":16},24]