The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides instructions for the agent to execute shell commands using pnpm and bash. These commands are used to run test suites (e.g., vitest), integration tests, and custom scripts (e.g., scripts/test-viewer-clipboard.sh) that are part of the repository's testing infrastructure.
[INDIRECT_PROMPT_INJECTION]: The skill defines workflows that ingest and process data from browser sessions, creating a potential surface for indirect prompt injection.
Ingestion points: The skill instructions involve reading and verifying browser session data, including cookies, localStorage, IndexedDB, service workers, and database records as described in SKILL.md.
Boundary markers: The instructions explicitly mandate the use of "disposable roots", "test accounts", and "disposable VM workflows" to provide isolation between the test data and the host environment.
Capability inventory: The skill possesses the capability to execute shell commands and scripts via pnpm and bash as defined throughout SKILL.md.
Sanitization: The skill relies on environment isolation (disposable VMs) as a primary security boundary rather than explicit string sanitization of the browser data.

verify-browser-sessions