The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted data, including source code files, package.json configurations, and CLI command outputs, which creates a surface for indirect prompt injection.
Ingestion points: The skill (SKILL.md) instructs the agent to read git diff, package.json exports, bin fields, and command outputs from the repository being audited.
Boundary markers: The instructions focus on locking the public surface but do not specify explicit delimiters or "ignore instructions" wrappers for the ingested content.
Capability inventory: The agent is authorized to read files, execute local build entry points (e.g., node ./dist/cli.js), and run packaging tools like npx publint.
Sanitization: The skill includes a mandatory safety instruction to "Inspect those scripts first or use a disposable checkout before calling --pack" to prevent the execution of potentially malicious lifecycle scripts (e.g., prepack).
[COMMAND_EXECUTION]: The skill performs local command execution of the software under audit (such as node ./dist/cli.js --help) to verify DX behavior and uses npx to run external diagnostic tools.
[EXTERNAL_DOWNLOADS]: The skill utilizes npx to fetch and execute the publint and @arethetypeswrong/cli packages from the npm registry for package validation.

dx-audit