The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes the !command syntax in references/ticket-sources.md to execute gh issue list during the loading or preprocessing phase. It also orchestrates multiple shell commands on the host including npx sandcastle init, pnpm sandcastle, and git push to manage the lifecycle of sandboxed agents and repository changes.
[PROMPT_INJECTION]: The skill's primary purpose is to ingest and process external ticket content from sources like GitHub issues and Linear. This untrusted data is interpolated into the prompts used by autonomous agents, creating a significant surface for indirect prompt injection attacks where malicious ticket descriptions or comments could influence the agent's behavior.
[CREDENTIALS_UNSAFE]: The skill requires high-privilege authentication tokens, specifically CLAUDE_CODE_OAUTH_TOKEN, GH_TOKEN, and LINEAR_API_KEY. The instructions suggest storing these in environment variables or .env files, which poses a risk of exposure if the host or the sandboxed environment is compromised.
[EXTERNAL_DOWNLOADS]: The skill relies on external tools and packages, including @ai-hero/sandcastle and tsx, which are downloaded and executed at runtime using npx and pnpm. These dependencies originate from outside the immediate skill environment and represent a supply chain risk.

sandcastle