The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and analyzes external content from GitHub release notes and remote changelog files to identify breaking changes and migration steps. This content is attacker-controlled (via the package repository) and could contain malicious instructions designed to influence the agent's behavior during the upgrade process.
Ingestion points: GitHub release bodies retrieved via gh release view and changelogs fetched via curl in references/release-notes.md.
Boundary markers: The skill does not specify the use of delimiters or instructions to ignore embedded prompts within the external text.
Capability inventory: The skill uses a wide range of powerful tools including bun, npm, gh, and git, and it executes verification gates (bun run typecheck, etc.) and code migrations (npx, bunx) in SKILL.md and references/migrations.md.
Sanitization: There is no evidence of sanitization or filtering of the external text before it is summarized for the agent.
[DYNAMIC_EXECUTION]: The skill utilizes dynamic execution of scripts and packages to perform audits and migrations.
Evidence: references/dependency-audit.md defines a script to be executed dynamically via bun -e to check for peer dependency conflicts.
Evidence: references/migrations.md encourages the use of package runners like npx, bunx, and pnpm dlx to execute codemods (e.g., @next/codemod) and transformation tools like jscodeshift directly from the registry.
[COMMAND_EXECUTION]: The skill's workflow relies on executing numerous shell commands for environment detection, dependency analysis, and project verification. This includes the use of find, rg (ripgrep), and jq to inspect the project structure and metadata.

deps-upgrade