Agents & models

Five agent roles exist, each backed by a different system prompt and tool set. The construction model is the simplest way to understand the split: one role drafts the blueprint, one keeps the job coherent, one performs trade work, one inspects, and one enforces deterministic checks.

Role	Responsibility	Typical model tier
spec	Surveys missing context and drafts the task blueprint in exploring.	High — reasoning-heavy.
coordinator	Keeps the project/domain plan coherent and makes promotion/rejection decisions.	High.
worker	Performs bounded trade work against the accepted blueprint.	Mid-to-high — coding-specific.
reviewer	Inspects completed work against the blueprint + rubric.	Mid.
gateChecker	Runs deterministic checks and custom gates.	Low — mostly shell/parse.

Assigning models

models:
  spec: claude-sonnet-4-6
  coordinator: claude-sonnet-4-6
  worker: qwen2.5-coder-32b-instruct
  reviewer: qwen2.5-coder-14b-instruct
  gateChecker: qwen2.5-coder-7b-instruct

Model IDs resolve against the catalog in ./src/core/models.ts.

Cognitive profiles

Each role has a CognitiveProfile declaring what it needs from a model — context window, tool use, reasoning. The coordinator uses these to validate model assignments at boot.

Permission modes

Agents run under a PermissionMode:

• plan — no side effects; agents can read but not write.
• default — standard allow-list; destructive operations prompt.
• full_auto — unrestricted; use only in sandboxed worktrees.

remediation_autonomy and worktree_isolation interact with permission mode to decide how aggressively agents may self-recover.

Persistence

Every agent conversation is snapshotted to ./memory/sessions/ per project. You can resume any agent from its last snapshot — see Sessions.