The Agent Skills Directory

[SAFE]: The skill operates within an isolated git worktree created specifically for the update process, ensuring that the primary working directory remains untouched and protected from accidental side effects during dependency experimentation.
[SAFE]: A strict 'Hard Contract' is defined that explicitly prohibits the agent from performing network-based side effects, such as executing git push or automatically opening pull requests, maintaining absolute user control over the remote repository state.
[COMMAND_EXECUTION]: Instructions guide the agent to use native package managers (including bun, npm, pnpm, yarn, go, uv, and cargo) for inventory and update tasks. These operations are restricted to the local environment and are standard for the skill's intended purpose of dependency maintenance.
[SAFE]: Indirect Prompt Injection Surface: The skill ingests external data including release notes and changelogs as documented in references/research-brief.md. Boundary markers are present in the form of a structured research template and risk classification system (references/risk-classification.md). Capabilities include standard file-write and shell execution for package updates. Sanitization is achieved through the use of native ecosystem tools rather than raw file manipulation.

batch-update