What problem does it solve?
Decides how to mutate a directed acyclic graph (DAG) when a node fails, output quality falls below threshold, or new information changes the plan, preventing wasted budget and repeated failures. It selects an appropriate recovery approach — retry, rephrase, upgrade/downgrade model, swap skills, fork paths, replan, insert validators, or escalate to humans — based on failure type, cost budget, and execution history. It complements error detection and execution components by focusing solely on strategy selection.
Core Features & Use Cases
- Failure-aware strategy selection: Maps failure types (transient, model, schema, logic, cascade) to targeted fixes like backoff retries, schema injection, or model upgrades.
- Cost-proportional recovery: Compares recovery cost vs remaining budget and avoids expensive recoveries for trivial failures.
- Iteration and escalation rules: Enforces iteration caps, traces root causes across dependencies, and escalates when automated strategies plateau.
- Parallel and fallback approaches: Supports forking parallel candidate paths to hedge ambiguity and swapping challenger skills when available.
- Integration points: Consumes diagnostics from dag-quality or dag-ops and hands off chosen plans to dag-planner or dag-runtime for execution.
Quick Start
Use the dag-mutation-strategist to pick a recovery plan for a failed DAG node by providing the failure type, execution history, and your cost budget.
