The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill is configured to read authentication tokens from the local file path ~/.config/vos/credentials to authorize API calls to https://vos.so/api. While this is the documented authentication method for the vendor's platform, accessing files outside the project directory is a sensitive operation.
[INDIRECT_PROMPT_INJECTION]: The skill fetches and processes remote .md files (recipes) from vos.so folders which are used to override the agent's default behavior and authoring rules.
Ingestion points: Remote recipe files (TASTE.md, DESIGN.md, etc.) and exemplar configurations fetched via the GET /api/folders/{id} endpoint.
Boundary markers: The instructions do not define boundary markers to separate instructions from data; instead, they explicitly state that recipes "override every default this skill or the authoring skill has."
Capability inventory: The skill possesses capabilities for shell command execution (vos check, vos push, etc.) and performs authenticated network writes (PUT, POST) to the vendor's API.
Sanitization: No sanitization of recipe content is mentioned before the instructions are interpolated into the agent's context for program generation.
[COMMAND_EXECUTION]: The skill uses the @vosjs/cli tool to perform operations such as vos check, vos still, and vos push, which involve executing shell commands on generated configuration files.
[DYNAMIC_EXECUTION]: The skill generates VosConfigJson (version 2) source code based on external recipes and subsequently uses CLI tools to compile, render, and validate this code locally.

vos-create