The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill fetches assets and source code from the official npm registry (registry.npmjs.org) and GitHub (github.com). These are established, well-known services, and the skill correctly implements integrity checks using SHA-256 hashes to verify the downloaded content.
[COMMAND_EXECUTION]: Provides patterns for executing shell commands such as cp, chmod, and substituteInPlace within the Nix build environment. These operations are performed within the Nix sandbox during the build process, which limits their access to the host system and ensures reproducible outputs.
[INDIRECT_PROMPT_INJECTION]: The skill defines a workflow that involves reading and interpreting external package.json files to determine binary paths and build scripts.
Ingestion points: The agent is instructed to read package.json in SKILL.md (Step 1: Identify package type and Step 4: Handle special cases) to locate entry points.
Boundary markers: Not explicitly defined for the file ingestion step.
Capability inventory: The resulting Nix derivations have the capability to execute build commands (bun run build) and modify file permissions (chmod +x).
Sanitization: The skill mitigates risks by explicitly instructing the use of npm_config_ignore_scripts=true and --ignore-scripts during dependency installation, preventing the execution of arbitrary lifecycle scripts from untrusted packages.

package-npm-nix