The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface because it instructs the agent to read and follow directions contained within repository files such as AGENTS.md, CLAUDE.md, and DESIGN.md. If these files contain malicious instructions, they could influence the agent's actions during the implementation phase.
Ingestion points: The agent reads AGENTS.md, CLAUDE.md, README, DESIGN.md, package.json, and the outputs of gic-app-warmup and supabase status.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to disregard embedded instructions within these files.
Capability inventory: The agent possesses the capability to execute shell commands (including package managers and infrastructure tools), read/write files in the /workspace, and perform Git operations (commit/push).
Sanitization: No sanitization or validation of the content within the repository files is performed before the agent acts on the information.
[COMMAND_EXECUTION]: The skill requires the agent to execute a variety of shell commands to set up the development environment and verify changes. This includes running package managers (npm, bun, pnpm, yarn), Docker commands, the Supabase CLI, and sandbox-specific tools like gic-app-warmup, gic-browser, and agent-browser. These commands are standard for the skill's intended purpose of managing a development workflow within a managed sandbox.

dev-workflow