The Agent Skills Directory

[SAFE]: The skill consists of documentation and instructions without any executable code, scripts, or hardcoded credentials. It does not perform unauthorized network requests or system modifications.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface by design, as it ingests external data to drive its orchestration logic.
Ingestion points: The agent is instructed to read workflow definitions from docs/WORKFLOWS.md and data/workflows.json in the local environment, and it also accepts user-provided project outcomes (SKILL.md).
Boundary markers: The instructions do not define boundary markers or provide the agent with guidance to ignore potentially malicious instructions embedded within the workflow data or user input.
Capability inventory: The skill has the ability to invoke other specialized skills, including those for browser automation (go-playwright) and security audits, which could be abused if the orchestration logic is compromised.
Sanitization: There is no defined process for validating or sanitizing the content retrieved from the workflow source files or user-provided objectives before they are used to influence the agent's behavior.

antigravity-workflows