The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external websites and local directories during its audits (e.g., HTML, SEO metadata, and network traffic). This provides an attack surface where malicious content in the audited source could attempt to influence the agent's summary or the resulting launch verdict.
Ingestion points: The skill reads external content via fetch (in scripts/lib/pages.mjs) and local files via readFileSync (in scripts/scan-secrets.mjs and scripts/lib/pages.mjs).
Capability inventory: The skill can write log files (.readyvibe/verification.md), execute the git binary via execFileSync, drive a headless browser via Playwright, and perform network requests to arbitrary URLs.
Sanitization: Findings are structured as data and summarized, but there is no explicit sanitization of the content extracted from target sites before it is presented to the agent.
Boundary markers: None identified for separating untrusted site content from the agent's instructions.
[COMMAND_EXECUTION]: The skill executes the git binary using node:child_process (execFileSync) in scripts/scan-secrets.mjs to identify tracked files and check for ignored secrets. The command is restricted to ls-files and check-ignore and operates on a user-provided root directory.
[DYNAMIC_EXECUTION]: The skill employs dynamic loading in scripts/lib/browser.mjs to optionally import playwright or playwright-core from the environment or local node_modules. While the package names are static, the paths are resolved at runtime using createRequire.resolve and pathToFileURL.

launch-verification