What problem does it solve?
The four core Forge Harness verification pipelines (harvest-loop, steel-quench, phantom-quench, sim-conductor) are designed to run independently with no automatic handoff between steps. Running them manually loses critical inter-step signal, such as a FAIL in the adversarial verification step not blocking the subsequent phantom claim check, leading to incomplete or misleading verification results.
Core Features & Use Cases
- Gated Sequential Execution: Automatically chains the four verification pipelines so each step's verdict controls whether the next step runs, eliminating manual coordination and missed signal.
- Flexible Execution Modes: Supports --full (complete 4-step sweep), --quick (fast adversarial + phantom check), and --no-sim (skip simulation step) modes to match different verification needs, from quick iteration checks to full pre-release gates.
- Aggregated Reporting & Resume: Compiles all findings into a single structured report with clear pass/fail/pending status, and supports resuming from failed steps without re-running already completed work.
- Use Case: Before merging a pull request to a shared forge-harness project, run a full pipeline sweep to catch design flaws, ungrounded claims, and simulation issues in one go, rather than running each verification tool separately and manually tracking results.
Quick Start
Use the pipeline-conductor skill to run a full end-to-end verification sweep of your forge-harness project before opening a pull request.