omnigent-agents

Verify an implementer's diff with an INDEPENDENT, different-vendor sub-agent (diff plus contract only); turn blocking issues into fix-tasks and loop until clean.

Run independent subtasks in parallel — one git worktree and one implementation sub-agent per task, each opening its own PR — then cross-review every PR. maestro never merges; the human does.

Have the Claude and GPT partners critique each other's answers across a configurable number of rounds (default 1) before converging on a synthesis. Use when the user wants the two perspectives stress-tested against each other, not just shown side by side.

Load per slice — the COORDINATOR's 3-role dispatch sequence (specifier → coder → refactorer, SEPARATE sessions, a handback to the coordinator between each) that drives ONE vertical slice from a failing acceptance test to refactored green in a single worktree. Load for every slice the wavefront dispatches.

Load ONCE at planning (coordinator + specifier) to turn a PRD into right-sized vertical slices tagged produces/consumes/touches, derive the dependency DAG, and tag the spine. Load before any slice is dispatched.

Load ONCE as the coordinator's orchestration loop after the DAG is built. Land the spine, then drive dependency-ordered waves of parallel independent slices until the DAG drains, and finish with the architect verification.

Load ONCE at the end — the final whole-feature join by a fresh architect at the integration HEAD. Run gates, mutation, and cross-slice DRY over the assembled feature and SIGN-OFF or BOUNCE, attributing each failure to its owning slice.

Load after each wave completes — owned by the coordinator/integrator, not the architect. Dup-scan the wave's worktrees to DETECT cross-slice duplication (routed to the owning slice, never rewritten inline), merge each slice branch into the integration branch one at a time re-greening the gates, and bounce anything beyond a mechanical merge-marker conflict to the owning slice's own session.

Delegate read-only investigation, debugging, audit, search, or code-understanding tasks to sub-agents and synthesize only from their structured reports. In bottega, load it at DAG construction (to ground real contract edges) and on a bounce (to attribute a failure to its owning slice).