The Agent Skills Directory

[SAFE]: The skill incorporates robust credential safety by explicitly instructing the agent to never request API keys in chat and to avoid printing or logging secrets like the endpoint_secret_key during account audits.
[SAFE]: All network requests and external documentation links target the official vendor domains (paddle.com).
[SAFE]: The skill includes logical guardrails to prevent the agent from making destructive changes (delete, archive, revoke) to live production entities, requiring explicit user approval for any writes.
[INDIRECT_PROMPT_INJECTION]: The skill reads data from the external Paddle API (products, catalog, account status) which serves as an ingestion point for potentially untrusted data. The risk is minimized by the use of structured data processing and specific field selection.
Ingestion points: Data returned from client.products.list, client.account.get, and other API endpoints in SKILL.md.
Boundary markers: Not present, but the agent is instructed to focus on specific resource fields.
Capability inventory: Arbitrary JavaScript execution within the MCP server's constraints via the execute tool.
Sanitization: Proactive removal of the endpoint_secret_key using jq filters and manual redaction logic in the audit scripts.

paddle-onboarding