zweihander

Use when completing tasks, implementing major features, or before merging to verify work meets requirements

Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes

Run a guarded commit-to-deploy workflow that verifies branch freshness, invokes commit and create-pr, waits for green CI, invokes approve-pr, monitors CD, and reports final deployment state. Use when the user asks to yolo-push, ship current changes, or execute the full PR-to-deployment flow.

Use when the user wants to analyze experiment results, inspect scores from a dataset run, check pass/fail rates, review per-item outputs, or deep-dive into experiment performance. Trigger phrases: "analyze results", "experiment scores", "how did the experiment perform", "show results", "inspect run", "experiment analysis".

Use when the user wants to compare two or more experiment runs, detect regressions, see score deltas between runs, or evaluate model performance differences. Trigger phrases include "compare runs", "compare experiments", "diff runs", "regression check", "which run is better", "model comparison", "A/B comparison".

This skill should be used when the user wants to configure a Langfuse dataset for remote experiment triggering from the UI, set up a webhook URL, update the default experiment payload, or enable the Custom Experiment feature. Trigger phrases include "configure remote experiment", "set webhook URL", "enable custom experiment", "set up experiment trigger", "configure dataset webhook".

This skill should be used when the user wants to create a new Langfuse dataset, set up a dataset for benchmarking, or create a dataset with input/output schema validation. Trigger phrases include "create dataset", "new dataset", "set up dataset", "add dataset".

Use when the user wants to create a Langfuse LLM-as-a-Judge evaluator. Trigger phrases: "create evaluator", "add evaluator", "new evaluation", "set up evaluation criteria", "create judge". Handles prompt composition, schema validation, ID generation, SQL insertion into eval_templates and job_configurations, and post-creation verification.

This skill should be used when the user wants to create a new Langfuse dashboard widget. Trigger phrases include "create widget", "add widget", "new visualization", "create chart", "add chart to dashboard". It handles ID generation, schema validation, SQL insertion, and post-creation verification of dashboard widgets.

Use when the user wants to delete or remove an LLM-as-a-Judge evaluator from Langfuse. Trigger phrases: "delete evaluator", "remove evaluator", "drop evaluator", "clean up evaluator". Handles safety checks, confirmation, transaction-wrapped deletion of job_configurations and eval_templates, and post-deletion verification.

Use when the user wants to delete/remove a Langfuse dashboard widget. Trigger phrases: "delete widget", "remove widget", "drop widget". Performs referential integrity checks against dashboards before deletion and cleans up dashboard references.

This skill should be used when the user needs help designing the input or expected output schema for Langfuse dataset items. This includes deciding what fields to include, how to structure item inputs for specific pipelines, and creating JSON Schema validation rules. Trigger phrases include "design schema", "item format", "what should my dataset items look like", "dataset item structure".

This skill should be used when the user asks to discover datasets, list datasets, find what datasets exist, browse dataset items, inspect dataset contents, check dataset schemas, or explore dataset metadata in their Langfuse project. It enumerates all datasets with their items, runs, and schema configurations.

Use this skill when the user needs to build or configure trace filters for LLM-as-a-Judge evaluators. It discovers all filterable dimensions in the project (trace names, tags, environments, observation names, levels), helps construct the filter JSON array, validates the filter, and optionally runs a count query to show how many traces match. Trigger phrases include "discover filters", "what can I filter on", "build filter", "configure evaluator filter", "show filter options".

This skill should be used when the user asks to discover models, list models, find what models are available, check model pricing, or see which models are actually used in their Langfuse project. It lists both the registered model definitions (with pricing) and the models observed in actual trace data via the Langfuse REST API with a direct-database fallback.

Use when the user asks to discover, list, enumerate, or inspect Langfuse score names, data types, or sources. Enumerates unique score names, data types, and sources via the Langfuse REST API with a direct-database fallback.

This skill should be used when the user asks to discover traces, list traces, find what traces exist, inspect trace names, explore trace tags, or check trace environments in their Langfuse project. It enumerates unique trace names, tags, environments, and user IDs via the Langfuse REST API with a direct-database fallback.

This skill should be used when the user asks to inspect an evaluator, view evaluator details, show evaluator configuration, check an evaluator's prompt, or see the full setup of a specific LLM-as-a-Judge evaluator. It shows the complete template, version history, and associated job configurations.

Use when calculating Langfuse dashboard widget grid positions. Trigger phrases: "layout widgets", "widget placement", "grid position", "arrange widgets", "widget positioning". Computes x/y coordinates and sizes for the 12-column dashboard grid.

Use when the user wants to list dataset runs, browse experiment runs, see what experiments have been executed, check run status, or inspect run metadata. Trigger phrases include "list runs", "show runs", "experiment runs", "dataset runs", "what experiments have been run", "browse runs".

This skill should be used when the user asks to list evaluators, see all evaluators, check what evaluators exist, or get an overview of LLM-as-a-Judge configurations in their Langfuse project. It queries eval_templates and job_configurations to produce a comprehensive summary.

This skill should be used when the user wants to list existing dashboard widgets in a Langfuse project, show widgets, see what widgets exist, get a widget inventory, or understand which widgets are deployed on which dashboards. Trigger phrases include "list widgets", "show widgets", "existing widgets", "what widgets exist", and "widget inventory". This is a shared skill used by both the data-explorer and widget-manager agents.

This skill should be used when the user wants to perform CRUD operations on Langfuse dashboards. Trigger phrases include "create dashboard", "list dashboards", "manage dashboards", "delete dashboard", "add widget to dashboard", "remove widget from dashboard". It handles listing, creating, updating metadata, deleting dashboards, and managing widget placement entries within dashboard definitions.

This skill should be used when the user wants to create, update, archive, delete, list, or bulk-populate dataset items in Langfuse. Trigger phrases include "add items", "create item", "populate dataset", "update item", "archive item", "delete item", "bulk import", "add test cases".

This skill should be used when the user wants to query metrics from Langfuse, run a metrics query, aggregate data, hit the v1 metrics API, or retrieve analytics data from Langfuse. Trigger phrases include "query metrics", "run metrics query", "metrics API", "v1 metrics", and "aggregate data".

This skill should be used when the user wants recommendations for meaningful Langfuse dashboard visualizations based on available data. Trigger phrases include "suggest widgets", "recommend visualizations", "what should I visualize", "widget suggestions", "dashboard recommendations". It discovers available data, analyzes patterns, and proposes relevant widget configurations.

This skill should be used when the user wants to activate, deactivate, pause, resume, enable, or disable an LLM-as-a-Judge evaluator in Langfuse. Trigger phrases include "activate evaluator", "deactivate evaluator", "pause evaluator", "enable evaluator", "disable all evaluators". It handles safety checks and supports bulk status changes.

This skill should be used when the user wants to trigger an experiment, run an experiment, start a dataset run, execute a benchmark, or send a webhook to start an experiment. Trigger phrases include "trigger experiment", "run experiment", "start experiment", "execute benchmark", "run dataset", "POST webhook".

Use when updating an existing Langfuse LLM-as-a-Judge evaluator. Trigger phrases include "update evaluator", "modify evaluator", "change evaluator prompt", "update evaluation filters", and "change evaluator model". Handles template-level changes (new version) and job-config-level changes (in-place update).

Use when the user wants to update or modify an existing Langfuse dashboard widget configuration. Triggers: "update widget", "modify widget", "change widget", "edit widget configuration". It reads the current widget state, accepts changes conversationally, validates the new configuration, and applies the UPDATE.

General-purpose Langfuse integration for observability, evaluations, datasets, dataset items, experiment runs, remote experiment webhooks, dashboard widgets, dashboards, metrics, traces, scores, models, and filter discovery. Use for any Langfuse task including discovering project data, querying metrics, creating or managing datasets, running or comparing experiments, creating or managing LLM-as-a-Judge evaluators, and creating or arranging dashboard visualizations.

Use when writing, editing, adding, reviewing, or planning Python unit and/or integration tests, mocks, fixtures, pytest structure, or test coverage.

Use when adapting skills for AI Assistant Harness Adaptation, using an assistant harness action matrix, or making explicitly named target skills harness-agnostic.

Audit and improve AGENTS.md files in repositories. Use when user asks to check, audit, update, improve, or fix AGENTS.md files. Scans for all AGENTS.md files, evaluates quality against templates, outputs quality report, then makes targeted updates. Also use when the user mentions "AGENTS.md maintenance" or "project memory optimization".

Use this skill when the user says "/ai-insights-hunter", "extract insights from conversation", "harvest this session", "what should we remember from this chat", "save insights from this conversation", or provides a path to an AI Assistant conversation log and wants to capture decisions, patterns, and preferences for future sessions. This skill reads a full conversation log (or finds it automatically), extracts valuable insights using parallel agents, interviews the user about what to keep, then stores the chosen insights in the right places to improve future AI Assistant interactions. Use PROACTIVELY when a long or complex session is wrapping up and the user wants to preserve learnings.

This skill should be used when the user asks to "audit my setup", "check my CLAUDE.md/AGENTS.md", "check my instruction files", "review my AI config", "clean up my instructions", "find conflicts in my setup", "optimize my agent setup", or mentions reviewing instruction files, skills folders, context files, or preferences for contradictions, redundancy, or dead weight.

Use when the user asks to audit, trim, compress, or reduce bloat, verbosity, redundancy, filler, or over-expanded vocabulary in Markdown loaded into AI agent or LLM context, especially skills, agent prompts, command prompts, and assistant instruction files.

Commit staged changes following Conventional Commits v1.0.0. Activate when user asks to commit, create a commit, or says /commit.

Create a pull request from the current branch. Activate when user asks to create a PR, open a pull request, submit for review, or says /create-pr.

This skill should be used when the user asks to "address PR comments", "fix PR feedback", "resolve PR threads", "handle review comments", "address review feedback", "respond to PR comments", "fix code review comments", "fix what reviewers said", "work on PR comments", "iterate on PR", or "rework PR". It fetches unresolved PR review threads, presents them for user confirmation, then makes code fixes, replies to threads, and resolves them.