The Agent Skills Directory

[DYNAMIC_EXECUTION]: The Argent flow system allows for the inclusion of local JavaScript modules (.mjs) using the script directive. This capability is intended for tasks like data seeding but enables arbitrary code execution within the environment where the flows are replayed.
[INDIRECT_PROMPT_INJECTION]: The skill generates executable YAML flows based on untrusted external data such as QA tickets, bug reports, and acceptance criteria.
Ingestion points: External test cases, tickets, and criteria documents processed by the agent to define automated flow steps in SKILL.md.
Boundary markers: The documentation does not specify the use of delimiters or 'ignore' instructions for ingested data in the YAML generation process.
Capability inventory: The skill has the ability to execute local scripts (.mjs), interact with physical and simulated mobile devices, and control Chromium-based applications through the Argent runner.
Sanitization: The skill employs a placeholder system ({{secret:NAME}}) for sensitive data, but its effectiveness is limited during the live recording phase where actual values are typed into the UI.
[DATA_EXFILTRATION]: There is a risk of exposing sensitive data due to how the recorder handles UI input fields. The 'Live authoring' documentation warns that typing secrets into non-secure fields (which are not automatically redacted) will return the plain text values to the agent's context, potentially leading to unauthorized data exposure to the LLM.

argent-create-flow