The Agent Skills Directory

[PROMPT_INJECTION]: Static analysis flagged potential concealment instructions. Evaluation of the content reveals these are benign UI/UX guidelines. The skill instructs the agent to track internal bookkeeping variables (such as session tokens and app IDs) silently to avoid cluttering the chat interface with technical metadata, which is a standard practice for maintaining a clean user experience.
[EXTERNAL_DOWNLOADS]: The skill performs automated fetches of documentation and version metadata using 'curl'. These downloads target official and well-known service domains including 'adapty.io', 'github.com', 'pub.dev', and 'maven.org'. These operations are performed neutrally to ensure the agent uses the most up-to-date SDK configurations.
[INDIRECT_PROMPT_INJECTION]: As the skill is designed to process external documentation at runtime, it includes a mandatory safety protocol in 'Phase 0'. This instruction explicitly commands the agent to treat all fetched content strictly as reference data and never as new instructions that could alter its behavior, permissions, or security posture.
[REMOTE_CODE_EXECUTION]: The skill installs the official Adapty CLI tool via 'npm' and executes standard platform build utilities such as 'gradlew', 'xcodebuild', and 'flutter analyze'. These actions are essential to the skill's primary purpose of implementing and verifying an SDK integration and occur within the scope of authorized developer workflows.
[DATA_EXFILTRATION]: While the skill tracks session markers for analytics, these are random tokens used solely for documentation grouping. There is no evidence of sensitive user data (such as credentials or private keys) being accessed or transmitted to external endpoints.

adapty-integration