event-description-generator

Orchestrates parallel subagent workers to audit and generate missing Amplitude event descriptions.

Updated Apr 3, 2026
One-click install
npx skills add https://github.com/amplitude/amplitude-copilot-plugin --skill event-description-generator-amplitude
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: event-description-generator
Source: https://github.com/amplitude/amplitude-copilot-plugin/tree/main/skills/plugins/amplitude-experimental/skills/event-description-generator
Command: npx skills add https://github.com/amplitude/amplitude-copilot-plugin --skill event-description-generator-amplitude

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes scripts (resource) and references (resource) components.

What problem does it solve? Auditing hundreds of Amplitude events for missing descriptions is slow and error-prone when done one event at a time. This Skill coordinates the entire audit at scale by dispatching bounded chunks of work to worker subagents and consolidating their results. ## Core Features & Use Cases - Orchestrated Chunk Processing: Resolves a project name to a projectId, plans cursor ranges, and spawns event-descriptions-worker subagents with bounded parallelism. - Run-Scoped Outputs: Writes per-chunk CSVs into run directories keyed by projectId and git SHA, then compresses them into a single CSV of suggested descriptions. - Controlled Write-Back: Optionally dispatches writer subagents to push approved descriptions to Amplitude via set_event_metadata, only when explicitly requested. - Use Case: A data team with 400 undocumented events runs the skill with cursorStart=0 and maxEvents=400, receives a compressed CSV of suggested descriptions, reviews it, then asks to write the descriptions back to Amplitude. ## Quick Start Ask the assistant to audit missing event descriptions for your Amplitude project starting at cursor 0 with a maximum of 400 events.

Frequently Asked Questions about event-description-generator

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I audit missing event descriptions in Amplitude at scale?▼

Provide a projectId or project name, then the skill plans cursor chunks and spawns event-descriptions-worker subagents in parallel. Results are written as per-chunk CSVs and compressed into one file containing only rows with suggested descriptions.

How do I write generated event descriptions back to Amplitude?▼

After a run completes, explicitly ask to write the descriptions. The skill runs a writer-plan script over the compressed CSV and spawns event-descriptions-writer subagents that call set_event_metadata in bounded parallel waves.

Can I run the event description audit without a projectId?▼

Yes, you can provide a project name instead. The skill calls get_context to resolve it to a single projectId, asks you to choose if multiple matches exist, and halts if neither name nor ID is provided.

What happens if the compress script fails during a run?▼

The skill notes the compression error in the final report but does not block the summary. Per-chunk worker CSVs remain available in the run directory, so you can inspect or retry compression manually.

What are the limits on parallel workers and chunk sizes?▼

Defaults are 50 events per worker, up to 4 parallel workers, and 400 max events per run. Writers use fixed constants of 100 events per writer and 4 parallel writers, which are not configurable.