Multi-Agent Orchestration
Assign separate agents bounded roles and define a result merge protocol.
Agent orchestration
Deduplication / Consensus
Merge findings by the same trigger and root cause while preserving distinctions.
Hub-and-Spoke SDLC State Machine
Let a coordinator own lifecycle state and accept verified transition artifacts.
Cost-Optimized Model Routing
Select a model using measured task requirements and an explicit escalation path.
Agent Memory Isolation
Separate private working state from approved shared handoff artifacts.
Adversarial Triad + Counterarguments
Separate finding, rebuttal and adjudication roles with evidence-based resolution.
Pull-Based Kanban Orchestration
Claim ready tasks atomically and respect ownership and prerequisites.
Fork vs Fresh Spawning
Choose inherited context for continuity and a fresh brief for independent judgment.
No-Pre-Judging Reviewer Dispatch
Give reviewers neutral evidence and scope without prescribing their verdict.
Batch Remediation: One Fixer for All Findings
Repair related findings as one owned change wave before scoped re-review.
Per-Task Interfaces Block (Consumes/Produces)
Declare each task's consumed and produced interfaces before separate execution.
Atomic claim with payload binding
Claim an action in one operation and bind its identity to the original payload.
Choose interaction by investigation shape
Choose independent reports for a known artifact and interacting investigators for competing causal hypotheses.
Separate workflow, perspective and entry point
Keep domain work in leaf roles and orchestration at the entry point that owns the decision.
Dependency-ready integration frontier
Run independent dependency-ready tasks in isolated contexts and serialize their integration.
Differentiated design briefs
Generate alternatives under intentionally different constraints instead of multiplying identical proposals.
Same-Artifact All-Reviewer Gate
When required by the workflow, require every independent reviewer to pass the same artifact against the same rubric.
Keep the Stop Supervisor Outside the Worker
Run the liveness deadline and stop decision in a separately controlled supervisor that remains available when the worker stalls.
Observe a Blocked Worker from a Separate Control Path
Observe a blocked task through an independent read-only path that does not depend on that task making progress.