The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill automates the extraction of secret names from the Lovable Cloud settings page by navigating to specific project URLs (e.g., lovable.dev/projects/ID?view=cloud). It uses browser automation to scrape identifiers for configured secrets like STRIPE_SECRET_KEY or RESEND_API_KEY. While the skill documentation claims to avoid extracting the secret values themselves, the automated access to and scraping of credential management interfaces represents a high-risk behavior targeting sensitive configuration infrastructure.
[DYNAMIC_EXECUTION]: The skill provides specific JavaScript code blocks to be executed in the user's browser via the javascript_tool. These snippets are used to perform direct DOM queries, detect page state (e.g., GitHub sync status), and extract data from the Lovable UI. Execution of provided code in a browser session that may contain active authentication cookies is a significant capability risk.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it incorporates unvalidated data from local project metadata into the prompts it sends to the external Lovable service.
Ingestion points: Git commit messages (via git log) and filenames within the supabase/functions/ and supabase/migrations/ directories.
Boundary markers: Absent. The skill directly interpolates these unvalidated strings into the final prompt (e.g., "Deploy the [function-name] edge function").
Capability inventory: The skill possesses extensive capabilities including browser navigation, form input, clicking, and MCP tool invocation (send_message).
Sanitization: None detected; the skill treats commit messages and filenames as trusted inputs for prompt generation.
[COMMAND_EXECUTION]: The skill executes local shell commands, specifically git utilities (git diff, git log, git status), to monitor project state and determine when to trigger its automated deployment workflows.

yolo