The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs the agent to install the vplan package globally from NPM (npm i -g vplan@latest) if it is not already present. It also mentions potential Chromium downloads via Playwright (npx playwright install chromium) for export tasks. These are external dependencies retrieved from public registries at runtime.
[COMMAND_EXECUTION]: The workflow relies on multiple shell commands executed through the vplan CLI, such as vplan check, vplan <file>.mdx, and vplan export. The default review mode starts a long-running interactive server that blocks the terminal and requires manual exit or timeout.
[INDIRECT_PROMPT_INJECTION]: The skill generates visual plans based on user-supplied details, creating a surface for indirect prompt injection where untrusted input could manipulate the plan's output.
Ingestion points: User-provided descriptions and requirements for plan generation as noted in the SKILL.md frontmatter and Workflow section.
Boundary markers: The skill instructs the agent to use backticks for literal MDX characters, but there are no explicit instructions for creating security boundaries or using delimiters to isolate user-provided content.
Capability inventory: The agent can perform package installations via npm and execute CLI commands as described in SKILL.md and references/static-exports.md.
Sanitization: No validation or sanitization mechanisms for user input are defined before the data is processed into the MDX structure.

visual-plan