omnibus▌

### Querying PostHog Signals - Use the `document_embeddings` table with mandatory filters for `model_name`, `product`, `document_type`, and a `timestamp` range. - Always deduplicate results using `argMax(column, inserted_at)` grouped by `document_id` to handle soft-deleted or updated records. - Perform semantic searches using `cosineDistance` combined with the `embedText` function for natural language queries.

### PostHog HogQL Query Reference - Access comprehensive schema documentation for core system models including cohorts, feature flags, error tracking, and data warehouse. - Utilize a library of optimized analytical query templates for trends, funnels, retention, user paths, and LLM trace analysis. - Reference HogQL-specific syntax, available functions, and extensions for advanced data manipulation and session replay analysis.

### Instrument Product Analytics - Detect project framework and install the appropriate PostHog SDK using the project's native package manager. - Initialize the SDK and implement event tracking for key user actions, ensuring server-side events are captured where applicable. - Use environment variables for API keys and add identify calls during authentication to maintain consistent user tracking.

### PostHog LLM Analytics Instrumentation - Detect the LLM stack and install the PostHog SDK if not already present, using environment variables for all configuration keys. - Instrument LLM calls to capture essential metrics including token counts, model names, latency, and costs for every generation. - Utilize provider-specific reference files for framework-accurate callbacks or fallback to manual capture for unsupported integrations.

### Cleaning up stale feature flags - Identify stale flags using the PostHog API, filtering for unused or fully rolled-out configurations that are at least 30 days old. - Assess candidates by checking for experiment associations and dependencies, then present a summary to the user for explicit approval. - Disable flags before deletion to ensure safety, then provide tailored code cleanup instructions based on the flag's rollout state.

### Auditing PostHog Experiments and Flags - Perform configuration audits using read_data and list_data to identify staleness, misconfigurations, and best-practice violations. - Categorize findings by severity (CRITICAL, WARNING, INFO) and provide actionable remediation steps for each identified issue. - Use inline markdown for single-entity checks and create notebook artifacts for comprehensive audits exceeding five entities.

### Analyzing Experiment Session Replays - Retrieve experiment metadata and variant keys from the feature flag to ensure accurate filtering for each test group. - Construct session recording filters using the $feature_flag_called event and the specific $feature/<flag_key> property for each variant. - Compare session duration, activity metrics, and navigation paths across variants to provide qualitative insights on user behavior.

### PostHog SDK Integration - Detect the project framework and language via dependency files to select the appropriate reference guide for SDK installation. - Initialize the SDK using environment variables for API keys and implement user identification during login and signup events. - Follow framework-specific patterns for provider setup and ensure code consistency by matching existing project naming conventions.

### Exploring LLM Clusters - Identify usage patterns by analyzing AI/LLM traffic clusters generated via embedding similarity and Temporal workflows. - Use SQL queries to retrieve cluster metadata, compute cost and latency metrics, and identify high-error or high-cost traffic segments. - Drill into specific traces using trace IDs to inspect representative samples and verify AI-generated cluster descriptions.

### PostHog Error Tracking Integration - Detect the project platform and framework to install the appropriate PostHog SDK and configure environment variables securely. - Enable exception autocapture for global monitoring and add manual capture calls to specific error boundaries and catch blocks. - Configure source map uploads for frontend and mobile projects to ensure stack traces resolve to readable, original source code.