The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill supports a script: directive in YAML test files to execute local .mjs scripts (e.g., for seeding data). This involves dynamic loading from relative paths.\n
Evidence: references/e2e-yaml.md defines the script directive and its usage for tasks like seeding data.\n
Risk: Execution of arbitrary local code from relative paths.\n
Mitigation: The skill enforces a "request-only" policy and restricts script execution over remote links.\n- [COMMAND_EXECUTION]: The skill provides mechanisms to launch and interact with applications on multiple platforms (iOS, Android, Chromium, Vega).\n
Evidence: references/e2e-yaml.md describes the launch directive which accepts bundle identifiers or application paths.\n
Risk: The agent can start any application specified in the test file. This is a core functionality of the end-to-end testing tool.\n- [INDIRECT_PROMPT_INJECTION]: As an E2E testing tool, the agent interacts with and reads data from external application interfaces (iOS, Android, Chromium). These interfaces could potentially contain malicious content intended to influence agent behavior.\n
Ingestion points: describe, screenshot, and runner tree reads from target applications.\n
Capability inventory: Includes launch, tap, type, script, and e2e-execute (subprocess-like capabilities).\n
Sanitization: The runner redacts sensitive values resolved from secrets.

argent-create-e2e