The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Fetches authentication configuration (OAuth consumer credentials) from an external S3 bucket (thegarth.s3.amazonaws.com) at runtime. This establishes a dependency on a third-party managed resource for the skill's core authentication flow.
[COMMAND_EXECUTION]: The 'Self-repair' section in SKILL.md provides instructions for the AI agent to autonomously execute shell commands, such as using 'uvx' to download and run external Python packages, and to perform in-place modification of the skill's source code (scripts/garmin.ts) if it detects API errors.
[CREDENTIALS_UNSAFE]: Caches sensitive session tokens (OAuth1 and OAuth2) in a local configuration file (~/.config/garmin-api/tokens.json). While the script sets restricted file permissions (0600), these cached tokens allow persistent, programmatic access to the user's personal health and fitness history.
[PROMPT_INJECTION]: The skill processes health and activity data fetched from Garmin APIs without explicit sanitization or instruction boundaries. This presents an indirect prompt injection surface where malicious strings embedded in user-controlled data fields (like workout names) could influence agent behavior.

garmin-connect