The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from Microsoft Outlook (emails, events, and contacts) which serves as an ingestion point for potentially adversarial instructions from external senders.
Ingestion points: Data is ingested through scripts like scripts/getMessage.ts, scripts/listMessages.ts, and scripts/getEvent.ts.
Boundary markers: The skill does not currently implement specific delimiters or explicit instructions to ignore adversarial content within the processed email bodies or event descriptions.
Capability inventory: The skill possesses extensive capabilities, including scripts/sendMail.ts, scripts/deleteMessage.ts, and scripts/updateEvent.ts, which could be abused if an agent obeys injected instructions.
Sanitization: All inputs are validated against strict Zod schemas defined in lib/schemas.ts, providing a baseline layer of structural validation.
[DYNAMIC_EXECUTION]: The skill uses a local entry point, cli.js, to dynamically resolve and execute its logic.
Evidence: The cli.js script uses node:child_process.spawnSync to execute either the compiled JavaScript or the TypeScript source based on the environment. This is a standard architectural pattern for this vendor's tools to ensure cross-runtime compatibility.
[EXTERNAL_DOWNLOADS]: The documentation provides instructions for users to install the tool and its dependencies from remote sources.
Evidence: README.md and SKILL.md instruct users to use npm install and npx to fetch the package and its dependencies. These operations target official package registries and the author's verified repositories.

microsoft-outlook