The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted application data and user-defined flow files to automate UI interactions.
Ingestion points: Application UI hierarchies are read via tools like describe and native-full-hierarchy (references/flow-yaml.md).
Boundary markers: Instructions mandate the use of stable selectors (ids, text) and verification gates (identity and readiness checks) to ensure correct execution (SKILL.md).
Capability inventory: The skill utilizes tools to launch apps, perform UI gestures, and execute scripts (flow-yaml.md).
Sanitization: A dedicated secret management system is described to prevent credential leakage in reports and logs (references/live-authoring.md).
[DYNAMIC_EXECUTION]: The skill allows the inclusion of local JavaScript modules within automation flows.
The script: directive in flow YAML files enables execution of .mjs scripts (flow-yaml.md).
The flow-add-script tool integrates these scripts during the recording phase (references/live-authoring.md).
[COMMAND_EXECUTION]: The skill uses specialized tools to manage the lifecycle of target applications and devices.
Commands like restart-app, launch, and stop-all-simulator-servers are essential for establishing a clean test environment (references/reliability-and-recovery.md).
Safety documentation advises on the scope of these commands to avoid unintended side effects on concurrent sessions.

argent-create-flow