The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to read and interpret various local project files, such as package.json, tsconfig.json, nest-cli.json, and docker-build.sh. These files represent untrusted data sources that could potentially contain malicious instructions or metadata aimed at influencing the agent's behavior during a project evaluation or upgrade task.
Ingestion points: Processes configuration files and metadata across the entire target repository (e.g., in references/migrations/pnpm-latest/commands.md and references/migrations/nestjs-latest/flow.md).
Boundary markers: While it includes instructions to verify project facts and obtain user confirmation for destructive changes (e.g., in SKILL.md and references/migrations/nestjs-latest/index.md), it does not implement formal delimiters or sanitization for all interpolated data.
Capability inventory: The skill utilizes powerful command execution tools including pnpm, npm, bash, and docker to perform modifications and validations.
Sanitization: No explicit sanitization or filtering of project file content is mentioned before it is processed by the agent.
[COMMAND_EXECUTION]: The skill's primary functionality relies on executing a wide range of shell commands to perform migrations and upgrades. For example, it uses pnpm add, pnpm exec, docker build, and prisma generate. While these are expected for the skill's stated purpose, they provide a broad attack surface if the agent is successfully targeted by indirect prompt injection.
[EXTERNAL_DOWNLOADS]: The skill defines several workflows for updating project dependencies and installing new tools (e.g., oxfmt, oxlint, vitest, nest). Although these downloads typically target standard public registries like npm, the installation of numerous third-party packages is a fundamental part of the skill's operation.

upgrade-project