The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill presents an attack surface where instructions embedded in project data could influence the agent.
Ingestion points: The skill reads project-specific configuration files including biome.json, eslint.config.*, oxlint.config.ts, and package.json (documented in SETUP-AND-MIGRATION.md).
Boundary markers: No explicit delimiters or instructions to ignore embedded commands are present when reading configuration data.
Capability inventory: The skill can execute subprocesses via bunx, npx, pnpm, and yarn, and performs filesystem mutations using ultracite fix, init, and upgrade commands (documented in SETUP-AND-MIGRATION.md).
Sanitization: There is no mention of sanitization, validation, or escaping of the content read from these external configuration files before they are processed by the tool or agent.
[COMMAND_EXECUTION]: The skill directs the agent to run various shell commands using bunx, npx, pnpm exec, and yarn to perform environment diagnostics, apply code fixes, and initialize configurations.
[EXTERNAL_DOWNLOADS]: The skill instructions involve fetching external packages at runtime from remote registries, specifically using the npx ultracite@latest upgrade command to obtain the latest version of the toolset.
[REMOTE_CODE_EXECUTION]: The upgrade workflow involves the dynamic download and execution of a remote CLI tool, which is a common but sensitive operation for maintaining toolchains.

ultracite