The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The file mobile-project-prompt.md contains a hardcoded Supabase API key and Authorization bearer token (sb_publishable_uyxE__Tjagj8Zb4oXDVgHw_VE7rt2_J) used to authenticate with a remote database for data synchronization.
[DATA_EXFILTRATION]: The skill is designed to synchronize sensitive user data, including body measurements (height, weight, proportions), personal photos, and wardrobe inventory, to a remote Supabase database (ocbjtobfmbnbcvkrfzuo.supabase.co). This involves sending private information to external infrastructure managed by the vendor.
[EXTERNAL_DOWNLOADS]: The skill instructs users to install an external CLI tool, vinted-client, from a remote GitHub repository (github.com/googlarz/vinted-client) to enable resell functionality. While this is a vendor-owned resource, it introduces an external code dependency.
[COMMAND_EXECUTION]: The skill frequently uses shell commands and external CLI tools, such as curl for fetching weather data and the vinted-client CLI for posting listings to Vinted. It also creates local scheduled tasks that execute complex sequences involving Chrome automation and file system operations.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface due to its features that ingest and process untrusted data from the web.
Ingestion points: references/monitoring.md (processes user-provided URLs for new arrival alerts) and references/import.md (extracts data from Zalando and Amazon order history pages).
Boundary markers: Absent; the skill does not use delimiters or instructions to ignore embedded commands within the ingested web content.
Capability inventory: The agent can create scheduled tasks, write to local data files (profile.json, inventory.json, monitoring_state.json), and send external notifications via iMessage or email.
Sanitization: No sanitization or validation of the ingested external content is performed before it is processed or stored.

fashion