Create and manage prompt snippets — reusable text blocks referenced inside config variation prompts. Keeps common instructions, personas, and guardrails consistent across multiple configs.
This skill guides an agent through creating and managing prompt snippets in LaunchDarkly's AgentControl product. Prompt snippets are named, versioned, project-level text blocks — personas, safety guardrails, output-format instructions, domain knowledge, or regulatory disclaimers — that can be referenced inside multiple AI config variation prompts so common instructions stay consistent across configs. When a snippet is updated a new version is created, and referencing variations can pick up the latest version to keep everything in sync. The problem it solves is prompt duplication and drift across many configs.
The skill operates entirely through the remotely hosted LaunchDarkly MCP server, requiring the tools `create-prompt-snippet`, `list-prompt-snippets`, and `get-prompt-snippet`, with optional `update-prompt-snippet`, `delete-prompt-snippet`, and `update-ai-config-variation`. It lays out a workflow: identify reusable text by reviewing existing configs and checking for duplicates, create snippets with a descriptive lowercase-hyphenated key, human-readable name, text, optional description, and tags, then verify creation by fetching the snippet and confirming it appears in the project listing at version 1. Core principles emphasize reusability (only snippet text used in two or more places), single responsibility per snippet, descriptive keys, tagging for discovery, and verifying references. It also covers updates (each producing a new version), edge cases such as key collisions and long text, and guidance to warn before deleting a snippet that configs still reference.
The target users are teams managing LaunchDarkly AI configs who want to centralize and version-control shared prompt fragments. It is a benign configuration-management helper: it operates through official LaunchDarkly MCP tooling, performs create/read/update/list operations on prompt text, and includes a safety check before destructive deletes.
Named, versioned, project-level text blocks — such as personas, guardrails, output formats, domain knowledge, or compliance text — that can be referenced inside multiple LaunchDarkly AI config variation prompts to keep shared instructions consistent.
The remotely hosted LaunchDarkly MCP server must be configured, exposing the required tools create-prompt-snippet, list-prompt-snippets, and get-prompt-snippet (plus optional update, delete, and update-ai-config-variation tools).
Only create a snippet when the text will be used in two or more places. Text unique to a single variation should go directly in the prompt, and per-request dynamic content should use template variables instead.
Snippets are versioned: a newly created snippet is version 1, and each update creates a new version. Config variations referencing the snippet can pick up the new version to stay in sync.
The skill instructs you to warn the user that referencing configs will lose the reference and to delete only via delete-prompt-snippet with explicit confirmation.
Quick Setup:
.claude/skills/Repository
launchdarkly/agent-skills