The Agent Skills Directory

Credential Management Practices: The skill provides instructions for managing Vercel OIDC tokens using the Vercel CLI. It specifically recommends moving sensitive files such as .env.local and the .vercel directory from the repository's .gitignore to the local .git/info/exclude file. This configuration is local to the user's environment and does not persist across repository clones, which creates a potential consideration for accidental credential exposure if the manual exclusion steps are missed by other contributors.
Indirect Prompt Injection Surface: The workflow involves 'researcher' agents gathering external information from papers and API documentation, which is then passed to implementation and design agents. This data flow creates a surface for indirect prompt injection where untrusted external content could influence the behavior of sub-agents.
Ingestion points: Data enters the context via research/<angle>.md files (containing external findings) and brief.md (containing user-provided instructions).
Boundary markers: The prompt templates for sub-agents do not explicitly demonstrate the use of strict delimiters or instructions to ignore embedded commands in external research data.
Capability inventory: The skill possesses significant capabilities, including executing shell commands via npx subharness, performing git worktree operations, and running pnpm install and pnpm build processes.
Sanitization: The instructions do not specify explicit sanitization or filtering of the content retrieved during the research phase before it is interpolated into subsequent prompts.
Command Execution: The skill relies on shell command execution via npx and pnpm exec to coordinate sub-agents and manage build processes. While this is the intended functionality for a development-focused skill, it represents a capability that allows for broad filesystem and network operations within the workspace.

vgpu-agent-flow