The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes design requirements and brainstorming input which are then used to generate implementation plans. While the skill includes human-in-the-loop approval gates, the processing of potentially untrusted data to generate code-related specs constitutes an indirect injection surface.
Ingestion points: Processes user dialogue in the terminal and browser-based click events recorded in state_dir/events.
Boundary markers: Uses a 'HARD-GATE' requiring explicit human approval before any implementation actions and validation after each design section.
Capability inventory: Executes shell scripts for server lifecycle management, runs a local Node.js server, and writes design documentation to the project's documentation folder.
Sanitization: Employs a 'Spec Self-Review' phase to check for requirements ambiguity and scope issues.
[COMMAND_EXECUTION]: The skill uses system commands to manage a local web server and launch the default web browser for the visual companion.
Evidence: The start-server.sh script invokes node server.cjs, and server.cjs uses child_process.execFile to call platform-specific browser launchers like open, xdg-open, or rundll32.exe. It also allows for an optional BRAINSTORM_OPEN_CMD environment variable to be used as a custom launcher.

brainstorming