The Agent Skills Directory

[COMMAND_EXECUTION]: The skill defines procedures for preparing and executing shell commands and CLI tools to perform technical spikes.
The SKILL.md file references the omh runtime workflow-artifact decision-prototype prepare command used for metadata preparation.
references/procedure.md outlines steps for preparing executor-neutral handoffs containing exact commands and expected observations.
[DYNAMIC_EXECUTION]: The skill facilitates the creation and execution of temporary code artifacts to test logic, state, and interaction hypotheses.
references/artifact-shapes.md describes creating modules and UI variants in languages such as Python, JavaScript, and Ruby to be executed by a driver script.
The workflow is specifically designed for "disposable, isolated experiments" where code is generated at runtime and intended to be discarded after observation.
[INDIRECT_PROMPT_INJECTION]: The workflow ingests external data (decision questions and measurement methods) which can influence the logic of generated code and experimental outputs.
Ingestion points: The decision question and measurement method inputs defined in references/procedure.md enter the agent's context to guide artifact creation.
Boundary markers: The skill mandates the use of strict "scratch boundaries" and "temporary worktrees" to isolate all generated content and prevent unauthorized file writes.
Capability inventory: The skill utilizes file-write operations, shell command preparation, and execution within the isolated workspace.
Sanitization: It employs mandatory "scope checks" and "budget isolation checks" to ensure artifacts are minimal and use synthetic fixtures by default, reducing the risk of malicious payload execution or data exposure.

omh-decision-prototype