The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted external data to guide the generation of new agent skills.
Ingestion points: The skill reads a local *-semantic-model.md file and accepts a "task in plain English" from the user as primary inputs during the workflow (Steps 1 and 2).
Boundary markers: The instructions do not specify the use of delimiters or specific "ignore embedded instructions" warnings when processing the content of the semantic model file.
Capability inventory: The agent has the capability to write files (the generated SKILL.md) and interact with the environment via use-semantius CLI tools.
Sanitization: There are no explicit instructions to sanitize or validate the content of the model file or user task before they are interpolated into the generated skill's logic.
[COMMAND_EXECUTION]: The skill's primary purpose is to generate executable logic for new agent skills, including Bash and Bun (TypeScript/JavaScript) scripts. It provides patterns for the agent to follow when creating these scripts to automate model-aware tasks.
[SAFE]: The skill includes explicit security guardrails for the generation process, specifically regarding Role-Based Access Control (RBAC). It instructs the agent to never promise bypasses for row-level security filters (select_rules) and to include visibility callouts for any read operations against scoped entities. It also mandates approval gates for all write operations in the generated skills.

semantius-agent-maker