dsh-agent-evolution

Configuration

Field	Default	Purpose
maxDepth	3	Absolute delegation-depth limit for experiment children.
maxTokens	inherited	Optional positive output-token limit when a tool call does not override it.
allowedPresets	system-trust roster	Optional non-empty preset-id allowlist. Explicit entries may include user-trust presets.
listToolName	agent_experiment_list_presets	Model-visible discovery tool name.
runToolName	agent_experiment_run	Model-visible single-run tool name.
compareToolName	agent_experiment_compare	Model-visible comparison tool name.

Without allowedPresets, discovery and execution admit only presets from roots configured with trust: system. With an explicit allowlist, every id is resolved when the plugin loads; unknown or broken entries prevent activation. A requested preset is resolved again before execution so deletion or damage fails before an Agent starts.

Selecting a preset can change the child's model-visible tools, skills, prompts, compaction, and other Agent-plane plugins. The child inherits the parent's model route unless provider and model are supplied together, inherits the parent's explicit sandbox override, and receives delegated approval: never when approval is composed. Preset authorization does not replace deployment review of the capabilities that each preset mounts.

Tool names must be non-empty and distinct. allowedPresets cannot be empty, contain blank ids, or contain duplicates. maxDepth, maxTokens, and per-call max_tokens reject invalid values during configuration or request validation.