The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill automatically parses untrusted content from the Lovable web interface (chat logs, sidebar messages, and deployment logs) to trigger further automated actions.
Ingestion points: Page content retrieved via read_page, find, and javascript_tool on the lovable.dev domain.
Boundary markers: Absent; the skill lacks delimiters or specific instructions to ignore embedded commands in the data retrieved from the UI.
Capability inventory: The skill can control the browser (computer tool), execute code (javascript_tool), and call external APIs via MCP (send_message).
Sanitization: No sanitization or validation of retrieved text is performed before it is used to determine the next step in the deployment flow.
[COMMAND_EXECUTION]: The skill uses local shell commands to inspect the state of the git repository.
Evidence: references/detection-logic.md and references/post-push-automation.md use git status, git log, and git diff to identify changed files and commit hashes.
[DATA_EXFILTRATION]: The skill automates the exposure of sensitive configuration metadata from a remote settings page.
Evidence: references/secrets-extraction.md directs the agent to scrape the Lovable Cloud settings UI specifically to list the names of environment secrets like STRIPE_SECRET_KEY and OPENAI_API_KEY.
[DYNAMIC_EXECUTION]: The skill provides the agent with specific JavaScript code to be executed dynamically in the browser environment.
Evidence: references/automation-workflows.md and references/secrets-extraction.md contain JavaScript snippets used with javascript_tool to interact with the DOM and extract data.

yolo