The Agent Skills Directory

[SAFE]: The skill is designed to structure complex user objectives into manageable phases (Plan, Build, Test, Ship). It includes built-in safety guardrails in the implementation playbook, such as requiring explicit user approval for destructive actions and ensuring authorization for security testing.
[INDIRECT_PROMPT_INJECTION]: The skill establishes an attack surface by reading workflow instructions from local files that could theoretically be manipulated by external processes.
Ingestion points: Workflow definitions are read from docs/WORKFLOWS.md and data/workflows.json as specified in SKILL.md.
Boundary markers: Absent; the agent is instructed to treat these files as the 'Source of Truth' without explicit delimiters or warnings to ignore embedded commands.
Capability inventory: The skill orchestrates the invocation of other specialized local skills and handles the generation of project artifacts (SKILL.md).
Sanitization: None detected; the skill assumes the integrity of the referenced workflow metadata and documentation.

antigravity-workflows