The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses explicit instructions to override the agent's internal training, stating 'Everything you know about the AI SDK is outdated or wrong' and requiring it to verify knowledge against external docs.
[COMMAND_EXECUTION]: The skill directs the agent to use shell commands such as curl to retrieve model lists from ai-gateway.vercel.sh and grep for searching local source code. It also instructs on using package managers like pnpm to install dependencies.
[EXTERNAL_DOWNLOADS]: The skill fetches documentation from ai-sdk.dev and model metadata from ai-gateway.vercel.sh, which are well-known services provided by the technology vendor.
[REMOTE_CODE_EXECUTION]: The skill suggests running npx @ai-sdk/devtools, which involves downloading and executing code from a remote package registry.
[PROMPT_INJECTION]: Indirect Injection Surface: The skill is configured to read and process documentation from external URLs and local filesystem paths that could be influenced by external actors.
Ingestion points: SKILL.md (instructions to fetch from ai-sdk.dev), node_modules/ai/docs/.
Boundary markers: Absent.
Capability inventory: Shell command execution (curl, pnpm, npx).
Sanitization: Absent.

ai-sdk