The Agent Skills Directory

[SAFE]: The skill implements a secure migration workflow using a five-stage process (Audit, Wrap, Tools, Tracking, Evals) that prioritizes user confirmation.
[EXTERNAL_DOWNLOADS]: The skill recommends installing official LaunchDarkly SDKs and well-known provider packages (such as OpenAI, Anthropic, and LangChain) via standard package managers. It also references official LaunchDarkly CHANGELOGs on GitHub for API verification, which are trusted sources.
[INDIRECT_PROMPT_INJECTION]: The skill has an ingestion surface for untrusted data as it scans user code and external prompt files (YAML, JSON, MD) during the 'Stage 1 Audit'. * Ingestion points: User source code and configuration files containing prompts or agent definitions. * Boundary markers: The workflow explicitly requires the user to confirm the audit results ('confirm', 'add', 'fix', 'stop') before any code mutation occurs. * Capability inventory: The skill performs file reads (audit) and instructs the agent to perform file writes (migration), package installations, and LaunchDarkly API interactions via sibling skills. * Sanitization: The skill provides specific instructions for converting hardcoded interpolation logic to Mustache template syntax, ensuring consistent and safe rendering across the platform.

migrate